From ef3469be15c495839209f5e06e67183ddc89c122 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Sun, 13 Sep 2026 16:26:20 +0300
Subject: [PATCH 01/17] Add CheaperInference provider
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

CheaperInference is an OpenAI-compatible gateway that serves each request
from one of several providers for the requested model, ranked by discount,
speed or a balance of both, at or below the model maker's list price. Adds the
provider entry plus 16 relayed models, all using base_model against existing
lab metadata with provider-side cost, cache rates, long-context tiers and
reasoning_options only.

Rates come from GET /v1/models (pricing_version sha256:d7fdd65d…, checked
2026-09-13T10:00:38Z).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 providers/cheaperinference/logo.svg           |  4 ++++
 .../models/claude-opus-5.toml                 | 13 ++++++++++++
 .../models/claude-sonnet-5.toml               | 19 ++++++++++++++++++
 .../models/deepseek-v4-flash-0731.toml        | 20 +++++++++++++++++++
 .../models/deepseek-v4.1-flash.toml           | 19 ++++++++++++++++++
 .../models/gemini-3.7-flash.toml              | 13 ++++++++++++
 .../models/glm-5.3-flash.toml                 | 16 +++++++++++++++
 .../cheaperinference/models/glm-5.3.toml      | 13 ++++++++++++
 .../cheaperinference/models/gpt-5-mini.toml   | 13 ++++++++++++
 .../cheaperinference/models/gpt-5.6-luna.toml | 20 +++++++++++++++++++
 .../cheaperinference/models/gpt-5.6-sol.toml  | 20 +++++++++++++++++++
 .../models/gpt-5.6-terra.toml                 | 20 +++++++++++++++++++
 .../cheaperinference/models/gpt-6-astra.toml  | 20 +++++++++++++++++++
 .../cheaperinference/models/gpt-oss-120b.toml | 13 ++++++++++++
 .../cheaperinference/models/grok-4.5.toml     | 16 +++++++++++++++
 .../cheaperinference/models/kimi-k3.toml      | 19 ++++++++++++++++++
 .../cheaperinference/models/qwen-3-8-max.toml | 16 +++++++++++++++
 providers/cheaperinference/provider.toml      |  5 +++++
 18 files changed, 279 insertions(+)
 create mode 100644 providers/cheaperinference/logo.svg
 create mode 100644 providers/cheaperinference/models/claude-opus-5.toml
 create mode 100644 providers/cheaperinference/models/claude-sonnet-5.toml
 create mode 100644 providers/cheaperinference/models/deepseek-v4-flash-0731.toml
 create mode 100644 providers/cheaperinference/models/deepseek-v4.1-flash.toml
 create mode 100644 providers/cheaperinference/models/gemini-3.7-flash.toml
 create mode 100644 providers/cheaperinference/models/glm-5.3-flash.toml
 create mode 100644 providers/cheaperinference/models/glm-5.3.toml
 create mode 100644 providers/cheaperinference/models/gpt-5-mini.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.6-luna.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.6-sol.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.6-terra.toml
 create mode 100644 providers/cheaperinference/models/gpt-6-astra.toml
 create mode 100644 providers/cheaperinference/models/gpt-oss-120b.toml
 create mode 100644 providers/cheaperinference/models/grok-4.5.toml
 create mode 100644 providers/cheaperinference/models/kimi-k3.toml
 create mode 100644 providers/cheaperinference/models/qwen-3-8-max.toml
 create mode 100644 providers/cheaperinference/provider.toml

diff --git a/providers/cheaperinference/logo.svg b/providers/cheaperinference/logo.svg
new file mode 100644
index 00000000000..36847516c91
--- /dev/null
+++ b/providers/cheaperinference/logo.svg
@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="currentColor">
+  <path d="M9.107 2.786H15.429V8.839H9.589a3.213 3.213 0 0 0 0.321 6.375H15.214L9.434 20.994A9.107 9.107 0 1 1 9.107 2.786Z"/>
+  <path d="M22.179 3.429H22.607a1.286 1.286 0 0 1 1.286 1.286V19.714a1.393 1.393 0 0 1 -1.393 1.393H17.464V8.143Z"/>
+</svg>
diff --git a/providers/cheaperinference/models/claude-opus-5.toml b/providers/cheaperinference/models/claude-opus-5.toml
new file mode 100644
index 00000000000..6cf0f48ba34
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-5.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
new file mode 100644
index 00000000000..42e66d89a62
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
+base_model = "anthropic/claude-sonnet-5"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 1.4
+output = 7.0
+cache_read = 0.14
+cache_write = 1.7
+
+[limit]
+output = 64000
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
new file mode 100644
index 00000000000..4f97706c0ca
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 57.60% off the model maker's list price ($0.076/$0.153 per 1M).
+base_model = "deepseek/deepseek-v4-flash-0731"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.032227
+output = 0.064454
+cache_read = 0.006445
+cache_write = 0.032227
+
+[limit]
+context = 1048576
+output = 65536
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
new file mode 100644
index 00000000000..2e1ccd3b2cd
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 59.72% off the model maker's list price ($0.3/$1.2 per 1M).
+base_model = "deepseek/deepseek-v4.1-flash"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.120853
+output = 0.483412
+cache_read = 0.002417
+cache_write = 0.120853
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/gemini-3.7-flash.toml b/providers/cheaperinference/models/gemini-3.7-flash.toml
new file mode 100644
index 00000000000..8e7cf605814
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.7-flash.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$3.75 per 1M).
+base_model = "google/gemini-3.7-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 0.525
+output = 2.625
+cache_read = 0.0525
+cache_write = 0.02919
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
new file mode 100644
index 00000000000..6c246456abe
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.5 per 1M).
+base_model = "zhipuai/glm-5.3-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.105
+output = 0.35
+cache_read = 0.01275
+cache_write = 0.105
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
new file mode 100644
index 00000000000..4753dcd6205
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
+base_model = "zhipuai/glm-5.3"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.77
+output = 2.42
+cache_read = 0.119
+cache_write = 0.77
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
new file mode 100644
index 00000000000..21416165b77
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$2.0 per 1M).
+base_model = "openai/gpt-5-mini"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.201421
+output = 1.611374
+cache_read = 0.020142
+cache_write = 0.201421
diff --git a/providers/cheaperinference/models/gpt-5.6-luna.toml b/providers/cheaperinference/models/gpt-5.6-luna.toml
new file mode 100644
index 00000000000..b201ca30969
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-luna.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($0.2/$1.2 per 1M).
+base_model = "openai/gpt-5.6-luna"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 0.08
+output = 0.48
+cache_read = 0.008
+cache_write = 0.1
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 0.16
+output = 0.72
+cache_read = 0.016
+cache_write = 0.2
diff --git a/providers/cheaperinference/models/gpt-5.6-sol.toml b/providers/cheaperinference/models/gpt-5.6-sol.toml
new file mode 100644
index 00000000000..ae6b0dc21c2
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-sol.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 50.00% off the model maker's list price ($2.0/$10.0 per 1M).
+base_model = "openai/gpt-5.6-sol"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 1.0
+output = 5.0
+cache_read = 0.1
+cache_write = 1.25
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 2.0
+output = 7.5
+cache_read = 0.2
+cache_write = 2.5
diff --git a/providers/cheaperinference/models/gpt-5.6-terra.toml b/providers/cheaperinference/models/gpt-5.6-terra.toml
new file mode 100644
index 00000000000..8ad8b8cde2f
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-terra.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($2.0/$12.0 per 1M).
+base_model = "openai/gpt-5.6-terra"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 0.8
+output = 4.8
+cache_read = 0.08
+cache_write = 1.0
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 1.6
+output = 7.2
+cache_read = 0.16
+cache_write = 2.0
diff --git a/providers/cheaperinference/models/gpt-6-astra.toml b/providers/cheaperinference/models/gpt-6-astra.toml
new file mode 100644
index 00000000000..7ddb4f8036e
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-6-astra.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "openai/gpt-6-astra"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 14.0
+output = 52.5
+cache_read = 1.4
+cache_write = 17.0
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
new file mode 100644
index 00000000000..396ea10c88f
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
+base_model = "openai/gpt-oss-120b"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 0.04
+output = 0.2
+cache_read = 0.0085
+cache_write = 0.04
diff --git a/providers/cheaperinference/models/grok-4.5.toml b/providers/cheaperinference/models/grok-4.5.toml
new file mode 100644
index 00000000000..90eb7fd10b8
--- /dev/null
+++ b/providers/cheaperinference/models/grok-4.5.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$6.0 per 1M).
+base_model = "xai/grok-4.5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 1.4
+output = 4.2
+cache_read = 0.17
+cache_write = 1.4
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
new file mode 100644
index 00000000000..f3b92762051
--- /dev/null
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "moonshotai/kimi-k3"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.239969
+cache_write = 2.1
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
new file mode 100644
index 00000000000..1c1d5d0e25f
--- /dev/null
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
+base_model = "alibaba/qwen3.8-max"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "xhigh"]
+
+[cost]
+input = 1.75
+output = 5.25
+cache_read = 0.2125
+cache_write = 2.125
diff --git a/providers/cheaperinference/provider.toml b/providers/cheaperinference/provider.toml
new file mode 100644
index 00000000000..b3bc7116068
--- /dev/null
+++ b/providers/cheaperinference/provider.toml
@@ -0,0 +1,5 @@
+name = "CheaperInference"
+env = ["CHEAPERINFERENCE_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://api.cheaperinference.com/v1"
+doc = "https://cheaperinference.com/docs"

From 702db21cd1b1a09548231e6e86f9360b158f6dac Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Sun, 13 Sep 2026 16:50:19 +0300
Subject: [PATCH 02/17] Drop toggle from CheaperInference reasoning options

The gateway documents that the reasoning field is forwarded to the serving
provider but names no on/off control of its own, so only the effort levels
inherited from the lab entries are authored.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 providers/cheaperinference/models/claude-sonnet-5.toml        | 3 ---
 providers/cheaperinference/models/deepseek-v4-flash-0731.toml | 3 ---
 providers/cheaperinference/models/deepseek-v4.1-flash.toml    | 3 ---
 providers/cheaperinference/models/kimi-k3.toml                | 3 ---
 providers/cheaperinference/models/qwen-3-8-max.toml           | 3 ---
 5 files changed, 15 deletions(-)

diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 42e66d89a62..3e1454e75c0 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "anthropic/claude-sonnet-5"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index 4f97706c0ca..ff083e943aa 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 57.60% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index 2e1ccd3b2cd..b3501817d28 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 59.72% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index f3b92762051..5d839270079 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index 1c1d5d0e25f..f9511623032 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "xhigh"]

From 8d7453b38bc88f0b2a503ee3c4d1e0ecd1c64f6b Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Sun, 13 Sep 2026 17:19:04 +0300
Subject: [PATCH 03/17] Restore toggle where the gateway forwards it

Tested on the live chat completions endpoint: reasoning.enabled = false
removes reasoning tokens and the reasoning field on kimi-k3, qwen-3-8-max,
deepseek-v4.1-flash and deepseek-v4-flash-0731, and true restores them, so
those four carry toggle with the wire path as a leading comment. On
claude-sonnet-5 neither reasoning.enabled nor thinking had any observable
effect on this path, so it stays effort-only.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 providers/cheaperinference/models/deepseek-v4-flash-0731.toml | 4 ++++
 providers/cheaperinference/models/deepseek-v4.1-flash.toml    | 4 ++++
 providers/cheaperinference/models/kimi-k3.toml                | 4 ++++
 providers/cheaperinference/models/qwen-3-8-max.toml           | 4 ++++
 4 files changed, 16 insertions(+)

diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index ff083e943aa..906b4220b64 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
 # Discounted marketplace rate, 57.60% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index b3501817d28..d94d5a4197f 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
 # Discounted marketplace rate, 59.72% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index 5d839270079..80e5d241f86 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index f9511623032..b8adf6c361e 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "xhigh"]

From 5f79a19e421ff8d55a0a92eef6e2ca07aedb2f7c Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Sun, 13 Sep 2026 17:21:15 +0300
Subject: [PATCH 04/17] Sync CheaperInference model catalog

Registers a sync module for the CheaperInference provider so the hourly
workflow keeps its rates current instead of leaving them to drift.

The gateway's catalog endpoint is authoritative for pricing: exact per-million
input, output, cache-read and cache-write rates, an above_threshold band with
its own token threshold for models priced in two bands, and the limits this
host actually serves. It exposes no reasoning controls, so the module sets
skipCreates and refuses to sync a reasoning model whose reasoning_options were
not hand-authored; image and video routes are skipped because their per-unit
media pricing cannot be expressed by the cost schema.

Reading the catalog needs a key, so the workflow gains a
CHEAPERINFERENCE_API_KEY secret. The endpoint accepts a key restricted to
read-only scope, which cannot spend from the account behind it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 .github/workflows/sync-models.yml             |   1 +
 packages/core/src/sync/index.ts               |   4 +
 .../src/sync/providers/cheaperinference.ts    | 191 ++++++++++++++++++
 packages/core/test/cheaperinference.test.ts   | 130 ++++++++++++
 4 files changed, 326 insertions(+)
 create mode 100644 packages/core/src/sync/providers/cheaperinference.ts
 create mode 100644 packages/core/test/cheaperinference.test.ts

diff --git a/.github/workflows/sync-models.yml b/.github/workflows/sync-models.yml
index ac9f3690bcd..3382485eb62 100644
--- a/.github/workflows/sync-models.yml
+++ b/.github/workflows/sync-models.yml
@@ -74,6 +74,7 @@ jobs:
           GH_TOKEN: ${{ github.token }}
           ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
           BASETEN_API_KEY: ${{ secrets.BASETEN_API_KEY }}
+          CHEAPERINFERENCE_API_KEY: ${{ secrets.CHEAPERINFERENCE_API_KEY }}
           DEEPINFRA_API_KEY: ${{ secrets.DEEPINFRA_API_KEY }}
           DIGITALOCEAN_API_TOKEN: ${{ secrets.DIGITALOCEAN_API_TOKEN }}
           DIGITALOCEAN_ACCESS_TOKEN: ${{ secrets.DIGITALOCEAN_ACCESS_TOKEN }}
diff --git a/packages/core/src/sync/index.ts b/packages/core/src/sync/index.ts
index 7f88aacb5bb..584b243f9fe 100644
--- a/packages/core/src/sync/index.ts
+++ b/packages/core/src/sync/index.ts
@@ -9,6 +9,7 @@ import { MissingReasoningOptionsError } from "./missing-reasoning-options.js";
 import { ambient } from "./providers/ambient.js";
 import { anthropic } from "./providers/anthropic.js";
 import { baseten } from "./providers/baseten.js";
+import { cheaperinference } from "./providers/cheaperinference.js";
 import { chutes } from "./providers/chutes.js";
 import { cloudflareAiGateway } from "./providers/cloudflare-ai-gateway.js";
 import { cloudflareWorkersAi } from "./providers/cloudflare-workers-ai.js";
@@ -133,6 +134,7 @@ export const providers: {
   ambient: SyncProvider<any>;
   anthropic: SyncProvider<any>;
   baseten: SyncProvider<any>;
+  cheaperinference: SyncProvider<any>;
   chutes: SyncProvider<any>;
   "cloudflare-ai-gateway": SyncProvider<any>;
   "cloudflare-workers-ai": SyncProvider<any>;
@@ -168,6 +170,7 @@ export const providers: {
   ambient,
   anthropic,
   baseten,
+  cheaperinference,
   chutes,
   "cloudflare-ai-gateway": cloudflareAiGateway,
   "cloudflare-workers-ai": cloudflareWorkersAi,
@@ -203,6 +206,7 @@ export const providers: {
 
 export const groups = {
   aggregators: [
+    "cheaperinference",
     "crossmodel",
     "edenai",
     "empiriolabs",
diff --git a/packages/core/src/sync/providers/cheaperinference.ts b/packages/core/src/sync/providers/cheaperinference.ts
new file mode 100644
index 00000000000..2f369d34b3a
--- /dev/null
+++ b/packages/core/src/sync/providers/cheaperinference.ts
@@ -0,0 +1,191 @@
+import { z } from "zod";
+
+import type { ExistingModel, SyncedFullModel, SyncedModel, SyncProvider } from "../index.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = "https://api.cheaperinference.com/v1/models";
+
+/** Long-context band. The gateway bills prompts above the threshold at these rates. */
+const AboveThreshold = z
+  .object({
+    input_token_price_threshold: z.number().int().positive(),
+    input_per_million: z.string().min(1),
+    output_per_million: z.string().min(1),
+    cache_read_input_per_million: z.string().min(1).nullish(),
+    cache_write_input_per_million: z.string().min(1).nullish(),
+  })
+  .passthrough();
+
+const CheaperInferencePricing = z
+  .object({
+    currency: z.literal("USD"),
+    input_per_million: z.string().min(1),
+    output_per_million: z.string().min(1),
+    cache_read_input_per_million: z.string().min(1).nullish(),
+    cache_write_input_per_million: z.string().min(1).nullish(),
+    input_token_price_threshold: z.number().int().positive().nullish(),
+    above_threshold: AboveThreshold.nullish(),
+  })
+  .passthrough();
+
+export const CheaperInferenceModel = z
+  .object({
+    id: z.string().min(1),
+    object: z.literal("model"),
+    type: z.string().min(1),
+    endpoint: z.string().min(1),
+    context_length: z.number().int().positive().nullish(),
+    max_output_tokens: z.number().int().positive().nullish(),
+    is_free: z.boolean(),
+    pricing: CheaperInferencePricing,
+  })
+  .passthrough();
+
+export const CheaperInferenceResponse = z
+  .object({
+    object: z.literal("list"),
+    data: z.array(CheaperInferenceModel),
+    pricing_version: z.string().min(1),
+    pricing_checked_at: z.string().min(1),
+  })
+  .passthrough();
+
+export type CheaperInferenceModel = z.infer<typeof CheaperInferenceModel>;
+
+export const cheaperinference = {
+  id: "cheaperinference",
+  name: "CheaperInference",
+  modelsDir: "providers/cheaperinference/models",
+  // The catalog carries pricing and limits but no reasoning controls, so new
+  // models need hand-authored reasoning_options before they can be created.
+  skipCreates: true,
+  sourceID(model) {
+    return isTokenPricedTextModel(model) ? model.id : undefined;
+  },
+  skippedNotice(ids) {
+    if (ids.length === 0) return [];
+    return [
+      `${ids.length} CheaperInference models were not created because the catalog exposes no reasoning controls, which this repo requires for reasoning models.`,
+      `Skipped remote IDs: ${ids.map((id) => `\`${id}\``).join(", ")}`,
+    ];
+  },
+  async fetchModels() {
+    return fetchCheaperInferenceModels();
+  },
+  parseModels(raw) {
+    return CheaperInferenceResponse.parse(raw).data;
+  },
+  translateModel(model, context) {
+    const existing = context.existing(model.id);
+    if (existing === undefined) return undefined;
+    return {
+      id: model.id,
+      model: buildCheaperInferenceModel(model, existing),
+    };
+  },
+} satisfies SyncProvider<CheaperInferenceModel>;
+
+export async function fetchCheaperInferenceModels(fetcher: typeof fetch = fetch) {
+  const apiKey = process.env["CHEAPERINFERENCE_API_KEY"];
+  if (apiKey === undefined || apiKey === "") {
+    throw new Error("CHEAPERINFERENCE_API_KEY is required to read the CheaperInference catalog");
+  }
+  const response = await fetcher(API_ENDPOINT, {
+    headers: { Authorization: `Bearer ${apiKey}` },
+  });
+  if (!response.ok) {
+    throw new Error(
+      `CheaperInference models request failed: ${response.status} ${response.statusText}`,
+    );
+  }
+  return CheaperInferenceResponse.parse(await response.json());
+}
+
+/**
+ * Only text models billed per token can be expressed by the catalog schema.
+ * Image and video routes price per unit of generated media, and free routes
+ * carry no rates worth syncing.
+ */
+function isTokenPricedTextModel(model: CheaperInferenceModel) {
+  return (
+    model.type === "text" &&
+    model.endpoint === "/v1/chat/completions" &&
+    !model.is_free &&
+    price(model.pricing.input_per_million) > 0 &&
+    price(model.pricing.output_per_million) > 0
+  );
+}
+
+function price(value: string) {
+  const parsed = Number(value);
+  if (!Number.isFinite(parsed) || parsed < 0) {
+    throw new Error(`CheaperInference returned an unusable price: ${value}`);
+  }
+  return parsed;
+}
+
+function optionalPrice(value: string | null | undefined) {
+  return value === null || value === undefined ? undefined : price(value);
+}
+
+export function buildCheaperInferenceModel(
+  model: CheaperInferenceModel,
+  existing: ExistingModel,
+): SyncedModel {
+  if (existing.reasoning !== false && existing.reasoning_options === undefined) {
+    throw new Error(
+      `CheaperInference model ${model.id} requires hand-authored reasoning_options; the catalog exposes no reasoning controls`,
+    );
+  }
+
+  const { base_model: baseModel, base_model_omit: baseModelOmit, ...current } = existing;
+  const pricing = model.pricing;
+
+  const cost = {
+    ...existing.cost,
+    input: price(pricing.input_per_million),
+    output: price(pricing.output_per_million),
+    cache_read: optionalPrice(pricing.cache_read_input_per_million),
+    cache_write: optionalPrice(pricing.cache_write_input_per_million),
+    tiers: buildTiers(pricing.above_threshold),
+  };
+
+  // The gateway is authoritative for what it serves, so its limits win when it
+  // publishes them. Missing values stay inherited from the lab entry.
+  const limit = {
+    ...existing.limit,
+    ...(model.context_length === null || model.context_length === undefined
+      ? {}
+      : { context: model.context_length }),
+    ...(model.max_output_tokens === null || model.max_output_tokens === undefined
+      ? {}
+      : { output: model.max_output_tokens }),
+  };
+
+  const values = {
+    ...current,
+    cost,
+    limit,
+  } as SyncedFullModel;
+
+  return baseModel === undefined
+    ? values
+    : factorBaseModel(baseModel, values, limit, baseModelOmit);
+}
+
+function buildTiers(above: z.infer<typeof AboveThreshold> | null | undefined) {
+  if (above === null || above === undefined) return undefined;
+  // The threshold is the last token billed at the base rate, so the band starts
+  // one token later. The gateway reports both 271_999 and 272_000 for the same
+  // 272k boundary, which normalises to one tier size.
+  const size = above.input_token_price_threshold + 1;
+  return [
+    {
+      tier: { type: "context" as const, size: size - (size % 1000) },
+      input: price(above.input_per_million),
+      output: price(above.output_per_million),
+      cache_read: optionalPrice(above.cache_read_input_per_million),
+      cache_write: optionalPrice(above.cache_write_input_per_million),
+    },
+  ];
+}
diff --git a/packages/core/test/cheaperinference.test.ts b/packages/core/test/cheaperinference.test.ts
new file mode 100644
index 00000000000..0f687650dfe
--- /dev/null
+++ b/packages/core/test/cheaperinference.test.ts
@@ -0,0 +1,130 @@
+import { expect, test } from "bun:test";
+
+import {
+  buildCheaperInferenceModel,
+  CheaperInferenceResponse,
+  cheaperinference,
+} from "../src/sync/providers/cheaperinference.js";
+
+function sourceModel(overrides: Record<string, unknown> = {}) {
+  return {
+    id: "claude-sonnet-5",
+    object: "model" as const,
+    type: "text",
+    endpoint: "/v1/chat/completions",
+    context_length: 1_000_000,
+    max_output_tokens: 64_000,
+    is_free: false,
+    pricing: {
+      currency: "USD" as const,
+      input_per_million: "1.400000",
+      output_per_million: "7.000000",
+      cache_read_input_per_million: "0.140000",
+      cache_write_input_per_million: "1.700000",
+      input_token_price_threshold: null,
+      above_threshold: null,
+    },
+    ...overrides,
+  };
+}
+
+const existing = {
+  base_model: "anthropic/claude-sonnet-5",
+  reasoning: true,
+  reasoning_options: [{ type: "effort" as const, values: ["low", "high"] }],
+  cost: { input: 1.4, output: 7 },
+  limit: { context: 1_000_000, output: 64_000 },
+};
+
+test("syncs the gateway's discounted rates, including cache rates", () => {
+  const model = buildCheaperInferenceModel(sourceModel() as never, existing as never);
+
+  expect(model).toMatchObject({
+    base_model: "anthropic/claude-sonnet-5",
+    cost: { input: 1.4, output: 7, cache_read: 0.14, cache_write: 1.7 },
+  });
+});
+
+test("turns a long-context band into a single context tier", () => {
+  const model = buildCheaperInferenceModel(
+    sourceModel({
+      id: "gpt-5.6-luna",
+      context_length: 1_050_000,
+      max_output_tokens: 128_000,
+      pricing: {
+        currency: "USD",
+        input_per_million: "0.080000",
+        output_per_million: "0.480000",
+        cache_read_input_per_million: "0.008000",
+        cache_write_input_per_million: "0.100000",
+        input_token_price_threshold: 271_999,
+        above_threshold: {
+          input_token_price_threshold: 271_999,
+          input_per_million: "0.160000",
+          output_per_million: "0.720000",
+          cache_read_input_per_million: "0.016000",
+          cache_write_input_per_million: "0.200000",
+        },
+      },
+    }) as never,
+    { ...existing, base_model: "openai/gpt-5.6-luna" } as never,
+  );
+
+  expect(model.cost?.tiers).toEqual([
+    {
+      tier: { type: "context", size: 272_000 },
+      input: 0.16,
+      output: 0.72,
+      cache_read: 0.016,
+      cache_write: 0.2,
+    },
+  ]);
+});
+
+test("takes the gateway's limits when it publishes them and inherits when it does not", () => {
+  const capped = buildCheaperInferenceModel(
+    sourceModel({ id: "deepseek-v4-flash-0731", context_length: 1_048_576, max_output_tokens: 65_536 }) as never,
+    { ...existing, base_model: "deepseek/deepseek-v4-flash-0731", limit: { context: 1_000_000, output: 384_000 } } as never,
+  );
+  expect(capped.limit).toMatchObject({ context: 1_048_576, output: 65_536 });
+
+  // With nothing published, the synced file authors no limit override at all
+  // and the lab entry's limits keep applying.
+  const unpublished = buildCheaperInferenceModel(
+    sourceModel({ id: "gpt-oss-120b", context_length: null, max_output_tokens: null }) as never,
+    { ...existing, base_model: "openai/gpt-oss-120b", limit: { context: 131_072, output: 32_768 } } as never,
+  );
+  expect(unpublished.limit).toBeUndefined();
+});
+
+test("refuses to sync a reasoning model that has no authored controls", () => {
+  expect(() =>
+    buildCheaperInferenceModel(sourceModel() as never, {
+      ...existing,
+      reasoning_options: undefined,
+    } as never),
+  ).toThrow(/reasoning_options/);
+});
+
+test("only syncs token-priced text routes", () => {
+  const ids = [
+    sourceModel(),
+    sourceModel({ id: "nano-banana-2", type: "image", endpoint: "/v1/images/generations" }),
+    sourceModel({ id: "seedance-2.0", type: "video", endpoint: "/v1/videos/generations" }),
+    sourceModel({ id: "some-free-model", is_free: true }),
+  ].map((model) => cheaperinference.sourceID(model as never));
+
+  expect(ids).toEqual(["claude-sonnet-5", undefined, undefined, undefined]);
+});
+
+test("parses the catalog response shape", () => {
+  const parsed = CheaperInferenceResponse.parse({
+    object: "list",
+    data: [sourceModel()],
+    pricing_version: "sha256:abc",
+    pricing_checked_at: "2026-09-10T21:01:09.012Z",
+    pricing_updated_at: "2026-08-05T00:00:00.000Z",
+  });
+
+  expect(parsed.data[0]?.id).toBe("claude-sonnet-5");
+});

From d99ca61ae3bc619036a48e3aac928b8fa80a331e Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 19:30:42 +0400
Subject: [PATCH 05/17] Refresh CheaperInference catalog from the live models
 endpoint

---
 .../models/claude-fable-5.1.toml              | 13 +++++++++++
 .../models/claude-fable-5.toml                | 13 +++++++++++
 .../models/claude-haiku-4.5.toml              | 11 ++++++++++
 .../models/claude-opus-4-8-fast.toml          | 12 ++++++++++
 .../models/claude-opus-4.5.toml               | 22 +++++++++++++++++++
 .../models/claude-opus-4.6.toml               | 18 +++++++++++++++
 .../models/claude-opus-4.7.toml               | 12 ++++++++++
 .../models/claude-opus-4.8.toml               | 15 +++++++++++++
 .../models/claude-opus-5-fast.toml            | 13 +++++++++++
 .../models/claude-opus-5.toml                 |  4 +++-
 .../models/claude-sonnet-4.5.toml             | 17 ++++++++++++++
 .../models/claude-sonnet-4.6.toml             | 18 +++++++++++++++
 .../models/claude-sonnet-5.toml               |  7 +++++-
 .../models/deepseek-v4-flash-0731.toml        | 18 ++++++---------
 .../models/deepseek-v4-flash.toml             | 16 ++++++++++++++
 .../models/deepseek-v4-pro-0813.toml          | 19 ++++++++++++++++
 .../models/deepseek-v4-pro.toml               | 16 ++++++++++++++
 .../models/deepseek-v4.1-flash.toml           | 13 +++++------
 .../models/gemini-2.5-flash.toml              | 14 ++++++++++++
 .../models/gemini-3-5-flash.toml              | 14 ++++++++++++
 .../models/gemini-3-flash-preview.toml        | 18 +++++++++++++++
 .../models/gemini-3.1-flash-lite.toml         | 11 ++++++++++
 .../models/gemini-3.1-pro-preview.toml        | 19 ++++++++++++++++
 .../models/gemini-3.6-flash.toml              | 13 +++++++++++
 .../models/gemini-3.7-flash.toml              |  2 +-
 .../cheaperinference/models/glm-4.5-air.toml  | 16 ++++++++++++++
 .../cheaperinference/models/glm-4.5.toml      | 16 ++++++++++++++
 .../cheaperinference/models/glm-4.6.toml      | 15 +++++++++++++
 .../cheaperinference/models/glm-4.7.toml      | 16 ++++++++++++++
 .../cheaperinference/models/glm-5.1.toml      | 15 +++++++++++++
 .../cheaperinference/models/glm-5.2.toml      | 16 ++++++++++++++
 .../models/glm-5.3-flash.toml                 | 10 ++++-----
 .../cheaperinference/models/glm-5.3.toml      |  2 +-
 providers/cheaperinference/models/glm-5.toml  | 16 ++++++++++++++
 .../cheaperinference/models/gpt-4.1-nano.toml |  9 ++++++++
 .../cheaperinference/models/gpt-5-mini.toml   |  4 +++-
 .../cheaperinference/models/gpt-5-nano.toml   | 15 +++++++++++++
 .../models/gpt-5.2-codex.toml                 | 19 ++++++++++++++++
 .../cheaperinference/models/gpt-5.4-mini.toml | 15 +++++++++++++
 .../cheaperinference/models/gpt-5.4-nano.toml | 13 +++++++++++
 .../cheaperinference/models/gpt-5.4.toml      | 19 ++++++++++++++++
 .../cheaperinference/models/gpt-5.5-pro.toml  | 16 ++++++++++++++
 .../cheaperinference/models/gpt-5.5.toml      | 16 ++++++++++++++
 .../cheaperinference/models/gpt-5.6-luna.toml |  4 ++--
 .../cheaperinference/models/gpt-5.6-sol.toml  |  4 ++--
 .../models/gpt-5.6-terra.toml                 |  4 ++--
 .../cheaperinference/models/gpt-6-astra.toml  |  2 +-
 .../cheaperinference/models/gpt-oss-120b.toml |  2 +-
 .../cheaperinference/models/grok-4.5.toml     |  2 +-
 .../cheaperinference/models/kimi-k3.toml      |  3 +--
 .../cheaperinference/models/minimax-m2.7.toml | 17 ++++++++++++++
 .../models/muse-spark-1.2.toml                | 16 ++++++++++++++
 .../cheaperinference/models/qwen-3-8-27b.toml | 16 ++++++++++++++
 .../cheaperinference/models/qwen-3-8-max.toml |  8 +++++--
 .../models/qwen3-5-35b-a3b.toml               | 20 +++++++++++++++++
 .../models/qwen3-6-35b-a3b.toml               | 18 +++++++++++++++
 .../cheaperinference/models/qwen3.6-27b.toml  | 19 ++++++++++++++++
 57 files changed, 690 insertions(+), 41 deletions(-)
 create mode 100644 providers/cheaperinference/models/claude-fable-5.1.toml
 create mode 100644 providers/cheaperinference/models/claude-fable-5.toml
 create mode 100644 providers/cheaperinference/models/claude-haiku-4.5.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-4-8-fast.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-4.5.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-4.6.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-4.7.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-4.8.toml
 create mode 100644 providers/cheaperinference/models/claude-opus-5-fast.toml
 create mode 100644 providers/cheaperinference/models/claude-sonnet-4.5.toml
 create mode 100644 providers/cheaperinference/models/claude-sonnet-4.6.toml
 create mode 100644 providers/cheaperinference/models/deepseek-v4-flash.toml
 create mode 100644 providers/cheaperinference/models/deepseek-v4-pro-0813.toml
 create mode 100644 providers/cheaperinference/models/deepseek-v4-pro.toml
 create mode 100644 providers/cheaperinference/models/gemini-2.5-flash.toml
 create mode 100644 providers/cheaperinference/models/gemini-3-5-flash.toml
 create mode 100644 providers/cheaperinference/models/gemini-3-flash-preview.toml
 create mode 100644 providers/cheaperinference/models/gemini-3.1-flash-lite.toml
 create mode 100644 providers/cheaperinference/models/gemini-3.1-pro-preview.toml
 create mode 100644 providers/cheaperinference/models/gemini-3.6-flash.toml
 create mode 100644 providers/cheaperinference/models/glm-4.5-air.toml
 create mode 100644 providers/cheaperinference/models/glm-4.5.toml
 create mode 100644 providers/cheaperinference/models/glm-4.6.toml
 create mode 100644 providers/cheaperinference/models/glm-4.7.toml
 create mode 100644 providers/cheaperinference/models/glm-5.1.toml
 create mode 100644 providers/cheaperinference/models/glm-5.2.toml
 create mode 100644 providers/cheaperinference/models/glm-5.toml
 create mode 100644 providers/cheaperinference/models/gpt-4.1-nano.toml
 create mode 100644 providers/cheaperinference/models/gpt-5-nano.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.2-codex.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.4-mini.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.4-nano.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.4.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.5-pro.toml
 create mode 100644 providers/cheaperinference/models/gpt-5.5.toml
 create mode 100644 providers/cheaperinference/models/minimax-m2.7.toml
 create mode 100644 providers/cheaperinference/models/muse-spark-1.2.toml
 create mode 100644 providers/cheaperinference/models/qwen-3-8-27b.toml
 create mode 100644 providers/cheaperinference/models/qwen3-5-35b-a3b.toml
 create mode 100644 providers/cheaperinference/models/qwen3-6-35b-a3b.toml
 create mode 100644 providers/cheaperinference/models/qwen3.6-27b.toml

diff --git a/providers/cheaperinference/models/claude-fable-5.1.toml b/providers/cheaperinference/models/claude-fable-5.1.toml
new file mode 100644
index 00000000000..c1bb0a0698f
--- /dev/null
+++ b/providers/cheaperinference/models/claude-fable-5.1.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-fable-5-1"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.175
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-fable-5.toml b/providers/cheaperinference/models/claude-fable-5.toml
new file mode 100644
index 00000000000..4051001834c
--- /dev/null
+++ b/providers/cheaperinference/models/claude-fable-5.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-fable-5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-haiku-4.5.toml b/providers/cheaperinference/models/claude-haiku-4.5.toml
new file mode 100644
index 00000000000..cf8bf0d4de2
--- /dev/null
+++ b/providers/cheaperinference/models/claude-haiku-4.5.toml
@@ -0,0 +1,11 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.0/$5.0 per 1M).
+base_model = "anthropic/claude-haiku-4-5"
+
+reasoning = false
+
+[cost]
+input = 0.7
+output = 3.5
+cache_read = 0.07
+cache_write = 0.85
diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
new file mode 100644
index 00000000000..91138d970d9
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -0,0 +1,12 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
+base_model = "anthropic/claude-opus-4-8"
+
+attachment = false
+reasoning = false
+
+[cost]
+input = 8.4
+output = 42.0
+cache_read = 0.84
+cache_write = 10.2
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
new file mode 100644
index 00000000000..1a7df15b9bf
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -0,0 +1,22 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-5"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
new file mode 100644
index 00000000000..b6380538793
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -0,0 +1,18 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-6"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "max"]
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-4.7.toml b/providers/cheaperinference/models/claude-opus-4.7.toml
new file mode 100644
index 00000000000..cb3064b2a02
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.7.toml
@@ -0,0 +1,12 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-7"
+
+attachment = false
+reasoning = false
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-4.8.toml b/providers/cheaperinference/models/claude-opus-4.8.toml
new file mode 100644
index 00000000000..821358da4f8
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.8.toml
@@ -0,0 +1,15 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-8"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-5-fast.toml b/providers/cheaperinference/models/claude-opus-5-fast.toml
new file mode 100644
index 00000000000..bd3ca1f35d0
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-5-fast.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-opus-5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-opus-5.toml b/providers/cheaperinference/models/claude-opus-5.toml
index 6cf0f48ba34..7f3684a606b 100644
--- a/providers/cheaperinference/models/claude-opus-5.toml
+++ b/providers/cheaperinference/models/claude-opus-5.toml
@@ -1,7 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-5"
 
+attachment = false
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
new file mode 100644
index 00000000000..c08b5badf98
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -0,0 +1,17 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "anthropic/claude-sonnet-4-5"
+
+attachment = false
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.21
+cache_write = 2.55
+
+[limit]
+context = 198000
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
new file mode 100644
index 00000000000..6011c3e0dfb
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -0,0 +1,18 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "anthropic/claude-sonnet-4-6"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "max"]
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.21
+cache_write = 2.55
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 3e1454e75c0..158e30e1d64 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -1,7 +1,12 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "anthropic/claude-sonnet-5"
 
+attachment = false
+
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index 906b4220b64..63c33962878 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -1,20 +1,16 @@
-# Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
-# Discounted marketplace rate, 57.60% off the model maker's list price ($0.076/$0.153 per 1M).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 40.00% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
-values = ["low", "high", "max"]
+values = ["none", "low", "high", "max"]
 
 [cost]
-input = 0.032227
-output = 0.064454
-cache_read = 0.006445
-cache_write = 0.032227
+input = 0.0456
+output = 0.0918
+cache_read = 0.00646
+cache_write = 0.0456
 
 [limit]
 context = 1048576
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
new file mode 100644
index 00000000000..60b7e342794
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
+base_model = "deepseek/deepseek-v4-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "high", "xhigh"]
+
+[cost]
+input = 0.071388
+output = 0.142777
+cache_read = 0.01105
+cache_write = 0.071388
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
new file mode 100644
index 00000000000..ec24161c2f0
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
+base_model = "deepseek/deepseek-v4-pro-0813"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "max"]
+
+[cost]
+input = 0.462
+output = 1.386
+cache_read = 0.01541
+cache_write = 0.462
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
new file mode 100644
index 00000000000..a4f90eab921
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
+base_model = "deepseek/deepseek-v4-pro"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "high", "xhigh"]
+
+[cost]
+input = 0.462
+output = 1.386
+cache_read = 0.038981
+cache_write = 0.462
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index d94d5a4197f..62275822a41 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -1,6 +1,5 @@
-# Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
-# Discounted marketplace rate, 59.72% off the model maker's list price ($0.3/$1.2 per 1M).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.6 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
 
 [[reasoning_options]]
@@ -11,10 +10,10 @@ type = "effort"
 values = ["low", "high", "max"]
 
 [cost]
-input = 0.120853
-output = 0.483412
-cache_read = 0.002417
-cache_write = 0.120853
+input = 0.105
+output = 0.42
+cache_read = 0.0021
+cache_write = 0.105
 
 [limit]
 context = 1048576
diff --git a/providers/cheaperinference/models/gemini-2.5-flash.toml b/providers/cheaperinference/models/gemini-2.5-flash.toml
new file mode 100644
index 00000000000..e7b20d51473
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-2.5-flash.toml
@@ -0,0 +1,14 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
+base_model = "google/gemini-2.5-flash"
+
+reasoning = false
+
+[cost]
+input = 0.21
+output = 1.75
+cache_read = 0.021
+cache_write = 0.05831
+
+[limit]
+output = 65535
diff --git a/providers/cheaperinference/models/gemini-3-5-flash.toml b/providers/cheaperinference/models/gemini-3-5-flash.toml
new file mode 100644
index 00000000000..1691853dee6
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3-5-flash.toml
@@ -0,0 +1,14 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.5/$9.0 per 1M).
+base_model = "google/gemini-3.5-flash"
+
+reasoning = false
+
+[cost]
+input = 1.05
+output = 6.3
+cache_read = 0.105
+cache_write = 0.05831
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/gemini-3-flash-preview.toml b/providers/cheaperinference/models/gemini-3-flash-preview.toml
new file mode 100644
index 00000000000..bf8da3e4f71
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3-flash-preview.toml
@@ -0,0 +1,18 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.5/$3.0 per 1M).
+base_model = "google/gemini-3-flash-preview"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.35
+output = 2.1
+cache_read = 0.035
+cache_write = 0.05831
+
+[limit]
+context = 256000
diff --git a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
new file mode 100644
index 00000000000..6e19627093c
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
@@ -0,0 +1,11 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$1.5 per 1M).
+base_model = "google/gemini-3.1-flash-lite"
+
+reasoning = false
+
+[cost]
+input = 0.201421
+output = 1.20853
+cache_read = 0.02125
+cache_write = 0.067113
diff --git a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
new file mode 100644
index 00000000000..f42ddb44228
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$12.0 per 1M).
+base_model = "google/gemini-3.1-pro-preview"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 1.4
+output = 8.4
+cache_read = 0.14
+cache_write = 0.2625
+
+[limit]
+context = 1000000
+output = 32768
diff --git a/providers/cheaperinference/models/gemini-3.6-flash.toml b/providers/cheaperinference/models/gemini-3.6-flash.toml
new file mode 100644
index 00000000000..f4b04ede6d6
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.6-flash.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 15.00% off the model maker's list price ($0.75/$3.75 per 1M).
+base_model = "google/gemini-3.6-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.6375
+output = 3.1875
+cache_read = 0.06375
+cache_write = 0.035416
diff --git a/providers/cheaperinference/models/gemini-3.7-flash.toml b/providers/cheaperinference/models/gemini-3.7-flash.toml
index 8e7cf605814..e975ecacf23 100644
--- a/providers/cheaperinference/models/gemini-3.7-flash.toml
+++ b/providers/cheaperinference/models/gemini-3.7-flash.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$3.75 per 1M).
 base_model = "google/gemini-3.7-flash"
 
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
new file mode 100644
index 00000000000..107ecb42422
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
+base_model = "zhipuai/glm-4.5-air"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.104739
+output = 0.605
+cache_read = 0.017
+cache_write = 0.104739
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
new file mode 100644
index 00000000000..5cd864a5492
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.5"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.33
+output = 1.21
+cache_read = 0.051
+cache_write = 0.33
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
new file mode 100644
index 00000000000..82aaa4e4fd3
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -0,0 +1,15 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.6"
+
+reasoning = false
+
+[cost]
+input = 0.33
+output = 1.21
+cache_read = 0.051
+cache_write = 0.33
+
+[limit]
+context = 198000
+output = 16384
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
new file mode 100644
index 00000000000..ac51a965b0e
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.7"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.322274
+output = 1.21
+cache_read = 0.051
+cache_write = 0.322274
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
new file mode 100644
index 00000000000..915a94452ed
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -0,0 +1,15 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
+base_model = "zhipuai/glm-5.1"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.77
+output = 2.42
+cache_read = 0.119
+cache_write = 0.77
+
+[limit]
+output = 24000
diff --git a/providers/cheaperinference/models/glm-5.2.toml b/providers/cheaperinference/models/glm-5.2.toml
new file mode 100644
index 00000000000..ad3609d303d
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.2.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.8/$2.55 per 1M).
+base_model = "zhipuai/glm-5.2"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "max"]
+
+[cost]
+input = 0.44
+output = 1.4025
+cache_read = 0.068
+cache_write = 0.44
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
index 6c246456abe..b382619c492 100644
--- a/providers/cheaperinference/models/glm-5.3-flash.toml
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -1,5 +1,5 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
-# Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.5 per 1M).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 46.29% off the model maker's list price ($0.15/$0.5 per 1M).
 base_model = "zhipuai/glm-5.3-flash"
 
 [[reasoning_options]]
@@ -7,10 +7,10 @@ type = "effort"
 values = ["low", "high", "max"]
 
 [cost]
-input = 0.105
-output = 0.35
+input = 0.080568
+output = 0.268535
 cache_read = 0.01275
-cache_write = 0.105
+cache_write = 0.080568
 
 [limit]
 context = 1048576
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
index 4753dcd6205..9c3d855d110 100644
--- a/providers/cheaperinference/models/glm-5.3.toml
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.3"
 
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
new file mode 100644
index 00000000000..f80ffc7f13b
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
+base_model = "zhipuai/glm-5"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.483412
+output = 1.546919
+cache_read = 0.085
+cache_write = 0.483412
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/gpt-4.1-nano.toml b/providers/cheaperinference/models/gpt-4.1-nano.toml
new file mode 100644
index 00000000000..b4ec41095b4
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-4.1-nano.toml
@@ -0,0 +1,9 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 15.00% off the model maker's list price ($0.1/$0.4 per 1M).
+base_model = "openai/gpt-4.1-nano"
+
+[cost]
+input = 0.085
+output = 0.34
+cache_read = 0.008499
+cache_write = 0.085
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
index 21416165b77..3e34216873f 100644
--- a/providers/cheaperinference/models/gpt-5-mini.toml
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -1,7 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$2.0 per 1M).
 base_model = "openai/gpt-5-mini"
 
+attachment = false
+
 [[reasoning_options]]
 type = "effort"
 values = ["minimal", "low", "medium", "high"]
diff --git a/providers/cheaperinference/models/gpt-5-nano.toml b/providers/cheaperinference/models/gpt-5-nano.toml
new file mode 100644
index 00000000000..179c285f58f
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5-nano.toml
@@ -0,0 +1,15 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.05/$0.4 per 1M).
+base_model = "openai/gpt-5-nano"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.040284
+output = 0.322274
+cache_read = 0.004028
+cache_write = 0.040284
diff --git a/providers/cheaperinference/models/gpt-5.2-codex.toml b/providers/cheaperinference/models/gpt-5.2-codex.toml
new file mode 100644
index 00000000000..b071ab80686
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.2-codex.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.75/$14.0 per 1M).
+base_model = "openai/gpt-5.2-codex"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh"]
+
+[cost]
+input = 1.225
+output = 9.8
+cache_read = 0.1225
+cache_write = 1.225
+
+[limit]
+context = 256000
+output = 65536
diff --git a/providers/cheaperinference/models/gpt-5.4-mini.toml b/providers/cheaperinference/models/gpt-5.4-mini.toml
new file mode 100644
index 00000000000..79f01368462
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4-mini.toml
@@ -0,0 +1,15 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$4.5 per 1M).
+base_model = "openai/gpt-5.4-mini"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 0.525
+output = 3.15
+cache_read = 0.0525
+cache_write = 0.525
diff --git a/providers/cheaperinference/models/gpt-5.4-nano.toml b/providers/cheaperinference/models/gpt-5.4-nano.toml
new file mode 100644
index 00000000000..2287a5d60ab
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4-nano.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.2/$1.25 per 1M).
+base_model = "openai/gpt-5.4-nano"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 0.14
+output = 0.875
+cache_read = 0.014
+cache_write = 0.14
diff --git a/providers/cheaperinference/models/gpt-5.4.toml b/providers/cheaperinference/models/gpt-5.4.toml
new file mode 100644
index 00000000000..01ace763f69
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$15.0 per 1M).
+base_model = "openai/gpt-5.4"
+
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 1.75
+output = 10.5
+cache_read = 0.175
+cache_write = 1.75
+
+[limit]
+context = 1000000
+output = 131072
diff --git a/providers/cheaperinference/models/gpt-5.5-pro.toml b/providers/cheaperinference/models/gpt-5.5-pro.toml
new file mode 100644
index 00000000000..ca07f180ee8
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.5-pro.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($30.0/$180.0 per 1M).
+base_model = "openai/gpt-5.5-pro"
+
+[[reasoning_options]]
+type = "effort"
+values = ["medium", "high", "xhigh"]
+
+[cost]
+input = 21.0
+output = 126.0
+cache_read = 2.55
+cache_write = 21.0
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/gpt-5.5.toml b/providers/cheaperinference/models/gpt-5.5.toml
new file mode 100644
index 00000000000..bb844b8055c
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.5.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$30.0 per 1M).
+base_model = "openai/gpt-5.5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 3.5
+output = 21.0
+cache_read = 0.35
+cache_write = 3.5
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/gpt-5.6-luna.toml b/providers/cheaperinference/models/gpt-5.6-luna.toml
index b201ca30969..efc57d4928c 100644
--- a/providers/cheaperinference/models/gpt-5.6-luna.toml
+++ b/providers/cheaperinference/models/gpt-5.6-luna.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.2/$1.2 per 1M).
 base_model = "openai/gpt-5.6-luna"
 
@@ -13,7 +13,7 @@ cache_read = 0.008
 cache_write = 0.1
 
 [[cost.tiers]]
-tier = { type = "context", size = 272_000 }
+tier = { type = "context", size = 271_999 }
 input = 0.16
 output = 0.72
 cache_read = 0.016
diff --git a/providers/cheaperinference/models/gpt-5.6-sol.toml b/providers/cheaperinference/models/gpt-5.6-sol.toml
index ae6b0dc21c2..e885c615e41 100644
--- a/providers/cheaperinference/models/gpt-5.6-sol.toml
+++ b/providers/cheaperinference/models/gpt-5.6-sol.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 50.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "openai/gpt-5.6-sol"
 
@@ -13,7 +13,7 @@ cache_read = 0.1
 cache_write = 1.25
 
 [[cost.tiers]]
-tier = { type = "context", size = 272_000 }
+tier = { type = "context", size = 271_999 }
 input = 2.0
 output = 7.5
 cache_read = 0.2
diff --git a/providers/cheaperinference/models/gpt-5.6-terra.toml b/providers/cheaperinference/models/gpt-5.6-terra.toml
index 8ad8b8cde2f..1a6d60adf09 100644
--- a/providers/cheaperinference/models/gpt-5.6-terra.toml
+++ b/providers/cheaperinference/models/gpt-5.6-terra.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($2.0/$12.0 per 1M).
 base_model = "openai/gpt-5.6-terra"
 
@@ -13,7 +13,7 @@ cache_read = 0.08
 cache_write = 1.0
 
 [[cost.tiers]]
-tier = { type = "context", size = 272_000 }
+tier = { type = "context", size = 271_999 }
 input = 1.6
 output = 7.2
 cache_read = 0.16
diff --git a/providers/cheaperinference/models/gpt-6-astra.toml b/providers/cheaperinference/models/gpt-6-astra.toml
index 7ddb4f8036e..cfa5462e77f 100644
--- a/providers/cheaperinference/models/gpt-6-astra.toml
+++ b/providers/cheaperinference/models/gpt-6-astra.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "openai/gpt-6-astra"
 
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
index 396ea10c88f..eca6ca5f23f 100644
--- a/providers/cheaperinference/models/gpt-oss-120b.toml
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
 base_model = "openai/gpt-oss-120b"
 
diff --git a/providers/cheaperinference/models/grok-4.5.toml b/providers/cheaperinference/models/grok-4.5.toml
index 90eb7fd10b8..96ae58cd9c6 100644
--- a/providers/cheaperinference/models/grok-4.5.toml
+++ b/providers/cheaperinference/models/grok-4.5.toml
@@ -1,4 +1,4 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$6.0 per 1M).
 base_model = "xai/grok-4.5"
 
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index 80e5d241f86..f5d123681cd 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -1,5 +1,4 @@
-# Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
 
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
new file mode 100644
index 00000000000..ba1cffd8281
--- /dev/null
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -0,0 +1,17 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
+base_model = "minimax/MiniMax-M2.7"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high"]
+
+[cost]
+input = 0.21
+output = 0.84
+cache_read = 0.0255
+cache_write = 0.255
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/muse-spark-1.2.toml b/providers/cheaperinference/models/muse-spark-1.2.toml
new file mode 100644
index 00000000000..f4adcb6ef8a
--- /dev/null
+++ b/providers/cheaperinference/models/muse-spark-1.2.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 15.00% off the model maker's list price ($1.25/$4.25 per 1M).
+base_model = "meta/muse-spark-1.2"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 1.0625
+output = 3.6125
+cache_read = 0.10625
+cache_write = 1.0625
+
+[limit]
+output = 943718
diff --git a/providers/cheaperinference/models/qwen-3-8-27b.toml b/providers/cheaperinference/models/qwen-3-8-27b.toml
new file mode 100644
index 00000000000..9426e7b0c8b
--- /dev/null
+++ b/providers/cheaperinference/models/qwen-3-8-27b.toml
@@ -0,0 +1,16 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.214/$2.55 per 1M).
+base_model = "alibaba/qwen3.8-27b"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "xhigh"]
+
+[cost]
+input = 0.1498
+output = 1.785
+cache_read = 0.01819
+cache_write = 0.1498
+
+[limit]
+output = 65536
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index b8adf6c361e..c3706a7a963 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -1,8 +1,9 @@
-# Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:d7fdd65d…, checked 2026-09-13T10:00:38Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
 
+attachment = false
+
 [[reasoning_options]]
 type = "toggle"
 
@@ -10,6 +11,9 @@ type = "toggle"
 type = "effort"
 values = ["low", "medium", "xhigh"]
 
+[[reasoning_options]]
+type = "budget_tokens"
+
 [cost]
 input = 1.75
 output = 5.25
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
new file mode 100644
index 00000000000..6fa13f12546
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -0,0 +1,20 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
+base_model = "alibaba/qwen3.5-35b-a3b"
+
+attachment = false
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 0.21875
+output = 0.875
+cache_read = 0.026562
+cache_write = 0.21875
+
+[limit]
+context = 256000
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
new file mode 100644
index 00000000000..c55093250e3
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -0,0 +1,18 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.1/$1.0 per 1M).
+base_model = "alibaba/qwen3.6-35b-a3b"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 0.07
+output = 0.7
+cache_read = 0.0085
+cache_write = 0.07
+
+[limit]
+output = 262144
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
new file mode 100644
index 00000000000..7dbabd0b4a3
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -0,0 +1,19 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 55.24% off the model maker's list price ($0.6/$3.6 per 1M).
+base_model = "alibaba/qwen3.6-27b"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "budget_tokens"
+
+[cost]
+input = 0.241706
+output = 1.611374
+cache_read = 0.024171
+cache_write = 0.241706
+
+[limit]
+context = 256000
+output = 32768

From 066556cabf1eb0f0ab9023130b0ae8a563ea14c6 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 19:31:40 +0400
Subject: [PATCH 06/17] Add the google/ prefixed route to the CheaperInference
 catalog

---
 .../models/google/gemini-3.5-flash-lite.toml        | 13 +++++++++++++
 1 file changed, 13 insertions(+)
 create mode 100644 providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml

diff --git a/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
new file mode 100644
index 00000000000..3cdced4c1ab
--- /dev/null
+++ b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
@@ -0,0 +1,13 @@
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
+base_model = "google/gemini-3.5-flash-lite"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.21
+output = 1.75
+cache_read = 0.021
+cache_write = 0.078333

From 66a455fb5d72d2d08ba1ddb80dbca0da1af0e447 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 19:52:06 +0400
Subject: [PATCH 07/17] Keep reasoning toggle only where a live request
 confirms the gateway forwards it

---
 providers/cheaperinference/models/claude-sonnet-5.toml      | 3 ---
 providers/cheaperinference/models/deepseek-v4-flash.toml    | 4 ++++
 providers/cheaperinference/models/deepseek-v4-pro-0813.toml | 1 +
 providers/cheaperinference/models/deepseek-v4-pro.toml      | 4 ++++
 providers/cheaperinference/models/deepseek-v4.1-flash.toml  | 1 +
 providers/cheaperinference/models/glm-4.5-air.toml          | 3 ---
 providers/cheaperinference/models/glm-4.5.toml              | 1 +
 providers/cheaperinference/models/glm-4.7.toml              | 1 +
 providers/cheaperinference/models/glm-5.1.toml              | 1 +
 providers/cheaperinference/models/glm-5.toml                | 1 +
 providers/cheaperinference/models/kimi-k3.toml              | 1 +
 providers/cheaperinference/models/qwen-3-8-max.toml         | 3 ---
 providers/cheaperinference/models/qwen3-5-35b-a3b.toml      | 1 +
 providers/cheaperinference/models/qwen3-6-35b-a3b.toml      | 1 +
 providers/cheaperinference/models/qwen3.6-27b.toml          | 1 +
 15 files changed, 18 insertions(+), 9 deletions(-)

diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 158e30e1d64..363827072d6 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -4,9 +4,6 @@ base_model = "anthropic/claude-sonnet-5"
 
 attachment = false
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
index 60b7e342794..1fe92e2666a 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
 base_model = "deepseek/deepseek-v4-flash"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["none", "high", "xhigh"]
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
index ec24161c2f0..99b8adf9f26 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro-0813"
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
index a4f90eab921..207a2da3c55 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -1,7 +1,11 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
 values = ["none", "high", "xhigh"]
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index 62275822a41..844b3b874ea 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.6 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
index 107ecb42422..b66996fe2d3 100644
--- a/providers/cheaperinference/models/glm-4.5-air.toml
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -2,9 +2,6 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
 base_model = "zhipuai/glm-4.5-air"
 
-[[reasoning_options]]
-type = "toggle"
-
 [cost]
 input = 0.104739
 output = 0.605
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
index 5cd864a5492..45aad806e7e 100644
--- a/providers/cheaperinference/models/glm-4.5.toml
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.5"
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
index ac51a965b0e..d3180d0bfb0 100644
--- a/providers/cheaperinference/models/glm-4.7.toml
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.7"
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
index 915a94452ed..51f3ddfd839 100644
--- a/providers/cheaperinference/models/glm-5.1.toml
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.1"
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
index f80ffc7f13b..12d4e21ed6e 100644
--- a/providers/cheaperinference/models/glm-5.toml
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
 base_model = "zhipuai/glm-5"
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index f5d123681cd..d346a7b616e 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index c3706a7a963..dd6956df802 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -4,9 +4,6 @@ base_model = "alibaba/qwen3.8-max"
 
 attachment = false
 
-[[reasoning_options]]
-type = "toggle"
-
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "xhigh"]
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
index 6fa13f12546..f5af5e2992b 100644
--- a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
 base_model = "alibaba/qwen3.5-35b-a3b"
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
index c55093250e3..0e125f5bf79 100644
--- a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.1/$1.0 per 1M).
 base_model = "alibaba/qwen3.6-35b-a3b"
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
index 7dbabd0b4a3..3cd84f20a80 100644
--- a/providers/cheaperinference/models/qwen3.6-27b.toml
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -1,3 +1,4 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 55.24% off the model maker's list price ($0.6/$3.6 per 1M).
 base_model = "alibaba/qwen3.6-27b"

From 3d184079bd6bd65b5f5d941b8e143bdb7c80d303 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 19:59:16 +0400
Subject: [PATCH 08/17] Restore the reasoning toggle for glm-4.5-air with its
 verified wire form

---
 providers/cheaperinference/models/glm-4.5-air.toml | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
index b66996fe2d3..5d55a21220f 100644
--- a/providers/cheaperinference/models/glm-4.5-air.toml
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -1,7 +1,11 @@
+# Toggle: thinking.type = enabled|disabled
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
 base_model = "zhipuai/glm-4.5-air"
 
+[[reasoning_options]]
+type = "toggle"
+
 [cost]
 input = 0.104739
 output = 0.605

From e71f1b586b9e45871fc4fc16f31543231b2bb9ce Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 21:30:26 +0400
Subject: [PATCH 09/17] Author reasoning controls from live gateway probes

Re-probed every model the review flagged on the live chat completions
endpoint, with a unique prompt per request so nothing came from a cache.

Reasoning comes back where the gateway catalog says it does not.
claude-opus-4-8-fast returns reasoning_content on every plain request
(7 of 7) and drops it for reasoning.enabled = false, and glm-4.6 returns
reasoning_content with 128 to 592 reasoning tokens for every graded effort
value. Both lose the reasoning = false line: opus-4-8-fast carries the
verified toggle plus the lab effort set, glm-4.6 the toggle its first-party
file also uses. gemini-3-5-flash reasons as well, and the gateway documents
its own set in a 400 - minimal, low, medium, high, thinking that cannot be
disabled - so it stays effort-only with those four values.

reasoning.enabled = false and reasoning.effort = none both switch reasoning
off on deepseek-v4-flash, deepseek-v4-pro and deepseek-v4-flash-0731, so
each keeps a single off control, the toggle, with graded values and no none:
high|xhigh, high|xhigh and low|high|max. minimax-m2.7 has a real toggle
(no reasoning in 4 of 4 requests with reasoning.enabled = false, reasoning
in every other form), so it replaces the empty options list.

The four Claude routes the review asked about - opus 4.5, opus 4.6, sonnet
4.5 and sonnet 4.6 - return no reasoning_content and no reasoning token
count at all, and effort, reasoning.enabled and thinking.budget_tokens are
each accepted with 200 and change nothing. budget_tokens goes, and
reasoning_options is empty rather than an effort list this host does not
honour. qwen-3-8-max, re-verified as the review asked: reasoning.enabled =
false no longer suppresses anything in 6 of 6 requests, so its toggle and
effort list go the same way.

The luna, sol and terra context tiers move from 271_999 to 272_000, the band
start the sync module emits and gpt-6-astra already carries.

claude-haiku-4.5, claude-opus-4.7, gemini-2.5-flash and gemini-3.1-flash-lite
keep reasoning = false: the catalog declares them non-reasoning and five wire
forms return nothing on each.

bun run validate passes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 .../cheaperinference/models/claude-opus-4-8-fast.toml    | 9 ++++++++-
 providers/cheaperinference/models/claude-opus-4.5.toml   | 8 +-------
 providers/cheaperinference/models/claude-opus-4.6.toml   | 8 +-------
 providers/cheaperinference/models/claude-sonnet-4.5.toml | 4 +---
 providers/cheaperinference/models/claude-sonnet-4.6.toml | 8 +-------
 .../cheaperinference/models/deepseek-v4-flash-0731.toml  | 6 +++++-
 providers/cheaperinference/models/deepseek-v4-flash.toml | 2 +-
 providers/cheaperinference/models/deepseek-v4-pro.toml   | 2 +-
 providers/cheaperinference/models/gemini-3-5-flash.toml  | 4 +++-
 providers/cheaperinference/models/glm-4.6.toml           | 4 +++-
 providers/cheaperinference/models/gpt-5.6-luna.toml      | 2 +-
 providers/cheaperinference/models/gpt-5.6-sol.toml       | 2 +-
 providers/cheaperinference/models/gpt-5.6-terra.toml     | 2 +-
 providers/cheaperinference/models/minimax-m2.7.toml      | 4 ++--
 providers/cheaperinference/models/qwen-3-8-max.toml      | 8 +-------
 15 files changed, 31 insertions(+), 42 deletions(-)

diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
index 91138d970d9..6326950b642 100644
--- a/providers/cheaperinference/models/claude-opus-4-8-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -1,9 +1,16 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
 
 attachment = false
-reasoning = false
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
 
 [cost]
 input = 8.4
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
index 1a7df15b9bf..1bbe26c4367 100644
--- a/providers/cheaperinference/models/claude-opus-4.5.toml
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -3,13 +3,7 @@
 base_model = "anthropic/claude-opus-4-5"
 
 attachment = false
-
-[[reasoning_options]]
-type = "effort"
-values = ["low", "medium", "high"]
-
-[[reasoning_options]]
-type = "budget_tokens"
+reasoning_options = []
 
 [cost]
 input = 3.5
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
index b6380538793..dd26588bc6d 100644
--- a/providers/cheaperinference/models/claude-opus-4.6.toml
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -3,13 +3,7 @@
 base_model = "anthropic/claude-opus-4-6"
 
 attachment = false
-
-[[reasoning_options]]
-type = "effort"
-values = ["low", "medium", "high", "max"]
-
-[[reasoning_options]]
-type = "budget_tokens"
+reasoning_options = []
 
 [cost]
 input = 3.5
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
index c08b5badf98..24c971eaa49 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -3,9 +3,7 @@
 base_model = "anthropic/claude-sonnet-4-5"
 
 attachment = false
-
-[[reasoning_options]]
-type = "budget_tokens"
+reasoning_options = []
 
 [cost]
 input = 2.1
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
index 6011c3e0dfb..3b2e72d95b4 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.6.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -3,13 +3,7 @@
 base_model = "anthropic/claude-sonnet-4-6"
 
 attachment = false
-
-[[reasoning_options]]
-type = "effort"
-values = ["low", "medium", "high", "max"]
-
-[[reasoning_options]]
-type = "budget_tokens"
+reasoning_options = []
 
 [cost]
 input = 2.1
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index 63c33962878..db0f2598960 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -1,10 +1,14 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 40.00% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
 
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
-values = ["none", "low", "high", "max"]
+values = ["low", "high", "max"]
 
 [cost]
 input = 0.0456
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
index 1fe92e2666a..8a90f294b0b 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -8,7 +8,7 @@ type = "toggle"
 
 [[reasoning_options]]
 type = "effort"
-values = ["none", "high", "xhigh"]
+values = ["high", "xhigh"]
 
 [cost]
 input = 0.071388
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
index 207a2da3c55..12150561530 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -8,7 +8,7 @@ type = "toggle"
 
 [[reasoning_options]]
 type = "effort"
-values = ["none", "high", "xhigh"]
+values = ["high", "xhigh"]
 
 [cost]
 input = 0.462
diff --git a/providers/cheaperinference/models/gemini-3-5-flash.toml b/providers/cheaperinference/models/gemini-3-5-flash.toml
index 1691853dee6..384c3717312 100644
--- a/providers/cheaperinference/models/gemini-3-5-flash.toml
+++ b/providers/cheaperinference/models/gemini-3-5-flash.toml
@@ -2,7 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.5/$9.0 per 1M).
 base_model = "google/gemini-3.5-flash"
 
-reasoning = false
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
 
 [cost]
 input = 1.05
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
index 82aaa4e4fd3..e447239e1ba 100644
--- a/providers/cheaperinference/models/glm-4.6.toml
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -1,8 +1,10 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.6"
 
-reasoning = false
+[[reasoning_options]]
+type = "toggle"
 
 [cost]
 input = 0.33
diff --git a/providers/cheaperinference/models/gpt-5.6-luna.toml b/providers/cheaperinference/models/gpt-5.6-luna.toml
index efc57d4928c..ea2f4c8af95 100644
--- a/providers/cheaperinference/models/gpt-5.6-luna.toml
+++ b/providers/cheaperinference/models/gpt-5.6-luna.toml
@@ -13,7 +13,7 @@ cache_read = 0.008
 cache_write = 0.1
 
 [[cost.tiers]]
-tier = { type = "context", size = 271_999 }
+tier = { type = "context", size = 272_000 }
 input = 0.16
 output = 0.72
 cache_read = 0.016
diff --git a/providers/cheaperinference/models/gpt-5.6-sol.toml b/providers/cheaperinference/models/gpt-5.6-sol.toml
index e885c615e41..f650d64b021 100644
--- a/providers/cheaperinference/models/gpt-5.6-sol.toml
+++ b/providers/cheaperinference/models/gpt-5.6-sol.toml
@@ -13,7 +13,7 @@ cache_read = 0.1
 cache_write = 1.25
 
 [[cost.tiers]]
-tier = { type = "context", size = 271_999 }
+tier = { type = "context", size = 272_000 }
 input = 2.0
 output = 7.5
 cache_read = 0.2
diff --git a/providers/cheaperinference/models/gpt-5.6-terra.toml b/providers/cheaperinference/models/gpt-5.6-terra.toml
index 1a6d60adf09..d6cc346b6dc 100644
--- a/providers/cheaperinference/models/gpt-5.6-terra.toml
+++ b/providers/cheaperinference/models/gpt-5.6-terra.toml
@@ -13,7 +13,7 @@ cache_read = 0.08
 cache_write = 1.0
 
 [[cost.tiers]]
-tier = { type = "context", size = 271_999 }
+tier = { type = "context", size = 272_000 }
 input = 1.6
 output = 7.2
 cache_read = 0.16
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
index ba1cffd8281..0417f4da7c7 100644
--- a/providers/cheaperinference/models/minimax-m2.7.toml
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -1,10 +1,10 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "minimax/MiniMax-M2.7"
 
 [[reasoning_options]]
-type = "effort"
-values = ["high"]
+type = "toggle"
 
 [cost]
 input = 0.21
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index dd6956df802..4bbd1e6d7a8 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -3,13 +3,7 @@
 base_model = "alibaba/qwen3.8-max"
 
 attachment = false
-
-[[reasoning_options]]
-type = "effort"
-values = ["low", "medium", "xhigh"]
-
-[[reasoning_options]]
-type = "budget_tokens"
+reasoning_options = []
 
 [cost]
 input = 1.75

From dca50c0df6ba3be874df9e44fffc2b460a38d70c Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 21:37:34 +0400
Subject: [PATCH 10/17] Drop budget_tokens where no reasoning budget field
 exists

qwen3-5-35b-a3b, qwen3-6-35b-a3b and qwen3.6-27b carried budget_tokens, but
this gateway has no reasoning-budget field on the OpenAI-compatible path:
reasoning.max_tokens = 64 and thinking.budget_tokens = 64 are both accepted
with 200 and the reply still carries around a thousand characters of
reasoning, the same as a plain request.

Their toggles are re-verified and stay: reasoning.enabled = false returns no
reasoning content on all three, and true restores it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 providers/cheaperinference/models/qwen3-5-35b-a3b.toml | 3 ---
 providers/cheaperinference/models/qwen3-6-35b-a3b.toml | 3 ---
 providers/cheaperinference/models/qwen3.6-27b.toml     | 3 ---
 3 files changed, 9 deletions(-)

diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
index f5af5e2992b..c8606655fef 100644
--- a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -8,9 +8,6 @@ attachment = false
 [[reasoning_options]]
 type = "toggle"
 
-[[reasoning_options]]
-type = "budget_tokens"
-
 [cost]
 input = 0.21875
 output = 0.875
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
index 0e125f5bf79..bb117766f63 100644
--- a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -6,9 +6,6 @@ base_model = "alibaba/qwen3.6-35b-a3b"
 [[reasoning_options]]
 type = "toggle"
 
-[[reasoning_options]]
-type = "budget_tokens"
-
 [cost]
 input = 0.07
 output = 0.7
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
index 3cd84f20a80..cfd7181f23d 100644
--- a/providers/cheaperinference/models/qwen3.6-27b.toml
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -6,9 +6,6 @@ base_model = "alibaba/qwen3.6-27b"
 [[reasoning_options]]
 type = "toggle"
 
-[[reasoning_options]]
-type = "budget_tokens"
-
 [cost]
 input = 0.241706
 output = 1.611374

From 6fdcfcbd714bd37b0c8b1b6c219c510b533429ca Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Wed, 16 Sep 2026 22:28:33 +0400
Subject: [PATCH 11/17] Author reasoning, attachments and side channels from
 live probes

Answers the third review round, all of it re-probed on the live endpoint with
a unique prompt per request.

reasoning = false replaces reasoning_options = [] on claude-opus-4.5,
claude-opus-4.6, claude-sonnet-4.5 and claude-sonnet-4.6: no request form -
plain, reasoning.enabled, reasoning.effort, thinking.budget_tokens - returns
reasoning_content or a reasoning token count on those routes, so an empty
options list would claim an always-on reasoner that is not there. The four
that already carried reasoning = false keep it and now cite the probe in a
leading comment. qwen-3-8-max keeps reasoning = true with an empty options
list, because it does return reasoning_content on every request and neither
the toggle nor effort = "none" suppresses it.

interleaved records the side channel where the reply carries one: 33 models
return reasoning_content, gemini-3-flash-preview returns reasoning_details,
and glm-4.7 returns its reasoning in a plain reasoning field, which the schema
covers as interleaved = true. The OpenAI-shaped routes count reasoning tokens
without exposing any content, so they carry no interleaved.

attachment is dropped on gemini-3-flash-preview and gemini-3.1-pro-preview,
which accept an image_url part; the other sixteen reject it (400, or every
upstream route failing) and now also override modalities.input to text only
so the two agree.

qwen-3-8-27b gets the toggle it actually forwards plus low|medium|xhigh and no
none. gpt-5.5-pro is responses-shaped: /v1/chat/completions returns 400 and
/v1/responses answers, where effort medium and high work.

bun run validate passes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 .../cheaperinference/models/claude-fable-5.1.toml    |  3 +++
 .../cheaperinference/models/claude-fable-5.toml      |  3 +++
 .../cheaperinference/models/claude-haiku-4.5.toml    |  4 +++-
 .../models/claude-opus-4-8-fast.toml                 | 10 +++++++++-
 .../cheaperinference/models/claude-opus-4.5.toml     | 12 ++++++++++--
 .../cheaperinference/models/claude-opus-4.6.toml     | 12 ++++++++++--
 .../cheaperinference/models/claude-opus-4.7.toml     | 10 +++++++++-
 .../cheaperinference/models/claude-opus-4.8.toml     | 10 +++++++++-
 .../cheaperinference/models/claude-opus-5-fast.toml  |  3 +++
 providers/cheaperinference/models/claude-opus-5.toml | 10 +++++++++-
 .../cheaperinference/models/claude-sonnet-4.5.toml   | 12 ++++++++++--
 .../cheaperinference/models/claude-sonnet-4.6.toml   | 12 ++++++++++--
 .../cheaperinference/models/claude-sonnet-5.toml     | 10 +++++++++-
 .../models/deepseek-v4-flash-0731.toml               |  3 +++
 .../cheaperinference/models/deepseek-v4-flash.toml   |  3 +++
 .../models/deepseek-v4-pro-0813.toml                 |  3 +++
 .../cheaperinference/models/deepseek-v4-pro.toml     |  3 +++
 .../cheaperinference/models/deepseek-v4.1-flash.toml |  3 +++
 .../cheaperinference/models/gemini-2.5-flash.toml    |  4 +++-
 .../cheaperinference/models/gemini-3-5-flash.toml    |  3 +++
 .../models/gemini-3-flash-preview.toml               |  3 ++-
 .../models/gemini-3.1-flash-lite.toml                |  4 +++-
 .../models/gemini-3.1-pro-preview.toml               |  3 ++-
 .../cheaperinference/models/gemini-3.7-flash.toml    |  3 +++
 providers/cheaperinference/models/glm-4.5-air.toml   |  3 +++
 providers/cheaperinference/models/glm-4.5.toml       |  3 +++
 providers/cheaperinference/models/glm-4.6.toml       |  3 +++
 providers/cheaperinference/models/glm-4.7.toml       |  1 +
 providers/cheaperinference/models/glm-5.1.toml       |  3 +++
 providers/cheaperinference/models/glm-5.2.toml       |  3 +++
 providers/cheaperinference/models/glm-5.3-flash.toml |  3 +++
 providers/cheaperinference/models/glm-5.3.toml       |  3 +++
 providers/cheaperinference/models/glm-5.toml         |  3 +++
 providers/cheaperinference/models/gpt-5-mini.toml    |  7 ++++++-
 providers/cheaperinference/models/gpt-5-nano.toml    |  7 ++++++-
 providers/cheaperinference/models/gpt-5.2-codex.toml | 10 +++++++++-
 providers/cheaperinference/models/gpt-5.4-mini.toml  |  7 ++++++-
 providers/cheaperinference/models/gpt-5.4.toml       |  7 ++++++-
 providers/cheaperinference/models/gpt-5.5-pro.toml   |  8 +++++++-
 providers/cheaperinference/models/gpt-oss-120b.toml  |  3 +++
 providers/cheaperinference/models/grok-4.5.toml      |  3 +++
 providers/cheaperinference/models/kimi-k3.toml       |  3 +++
 providers/cheaperinference/models/minimax-m2.7.toml  |  3 +++
 providers/cheaperinference/models/qwen-3-8-27b.toml  |  9 ++++++++-
 providers/cheaperinference/models/qwen-3-8-max.toml  | 12 +++++++++++-
 .../cheaperinference/models/qwen3-5-35b-a3b.toml     | 10 +++++++++-
 .../cheaperinference/models/qwen3-6-35b-a3b.toml     |  3 +++
 providers/cheaperinference/models/qwen3.6-27b.toml   |  3 +++
 48 files changed, 239 insertions(+), 27 deletions(-)

diff --git a/providers/cheaperinference/models/claude-fable-5.1.toml b/providers/cheaperinference/models/claude-fable-5.1.toml
index c1bb0a0698f..123d05632d5 100644
--- a/providers/cheaperinference/models/claude-fable-5.1.toml
+++ b/providers/cheaperinference/models/claude-fable-5.1.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5-1"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/claude-fable-5.toml b/providers/cheaperinference/models/claude-fable-5.toml
index 4051001834c..f492eabf1d8 100644
--- a/providers/cheaperinference/models/claude-fable-5.toml
+++ b/providers/cheaperinference/models/claude-fable-5.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/claude-haiku-4.5.toml b/providers/cheaperinference/models/claude-haiku-4.5.toml
index cf8bf0d4de2..216949b787c 100644
--- a/providers/cheaperinference/models/claude-haiku-4.5.toml
+++ b/providers/cheaperinference/models/claude-haiku-4.5.toml
@@ -1,7 +1,9 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.0/$5.0 per 1M).
 base_model = "anthropic/claude-haiku-4-5"
-
 reasoning = false
 
 [cost]
diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
index 6326950b642..cddfc43a14d 100644
--- a/providers/cheaperinference/models/claude-opus-4-8-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -1,10 +1,14 @@
 # Toggle: reasoning.enabled = true|false
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
@@ -17,3 +21,7 @@ input = 8.4
 output = 42.0
 cache_read = 0.84
 cache_write = 10.2
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
index 1bbe26c4367..2942f72ca24 100644
--- a/providers/cheaperinference/models/claude-opus-4.5.toml
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -1,9 +1,13 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-5"
-
 attachment = false
-reasoning_options = []
+reasoning = false
 
 [cost]
 input = 3.5
@@ -14,3 +18,7 @@ cache_write = 4.25
 [limit]
 context = 198000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
index dd26588bc6d..519be6b1f9f 100644
--- a/providers/cheaperinference/models/claude-opus-4.6.toml
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -1,12 +1,20 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-6"
-
 attachment = false
-reasoning_options = []
+reasoning = false
 
 [cost]
 input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.7.toml b/providers/cheaperinference/models/claude-opus-4.7.toml
index cb3064b2a02..a820ea31c7c 100644
--- a/providers/cheaperinference/models/claude-opus-4.7.toml
+++ b/providers/cheaperinference/models/claude-opus-4.7.toml
@@ -1,7 +1,11 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-7"
-
 attachment = false
 reasoning = false
 
@@ -10,3 +14,7 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.8.toml b/providers/cheaperinference/models/claude-opus-4.8.toml
index 821358da4f8..fe8debe0a3a 100644
--- a/providers/cheaperinference/models/claude-opus-4.8.toml
+++ b/providers/cheaperinference/models/claude-opus-4.8.toml
@@ -1,9 +1,13 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
@@ -13,3 +17,7 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-5-fast.toml b/providers/cheaperinference/models/claude-opus-5-fast.toml
index bd3ca1f35d0..9b2cbe78529 100644
--- a/providers/cheaperinference/models/claude-opus-5-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-5-fast.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-opus-5"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
diff --git a/providers/cheaperinference/models/claude-opus-5.toml b/providers/cheaperinference/models/claude-opus-5.toml
index 7f3684a606b..c83539c9b82 100644
--- a/providers/cheaperinference/models/claude-opus-5.toml
+++ b/providers/cheaperinference/models/claude-opus-5.toml
@@ -1,9 +1,13 @@
+# No image input on this host: an image_url part was rejected on two of three requests and the catalog reports
+# capabilities.vision false, so image input is not dependable here (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-5"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
@@ -13,3 +17,7 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
index 24c971eaa49..64d697cf707 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -1,9 +1,13 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-5"
-
 attachment = false
-reasoning_options = []
+reasoning = false
 
 [cost]
 input = 2.1
@@ -13,3 +17,7 @@ cache_write = 2.55
 
 [limit]
 context = 198000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
index 3b2e72d95b4..cde1f93590d 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.6.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -1,12 +1,20 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-6"
-
 attachment = false
-reasoning_options = []
+reasoning = false
 
 [cost]
 input = 2.1
 output = 10.5
 cache_read = 0.21
 cache_write = 2.55
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 363827072d6..1adde594198 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -1,9 +1,13 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "anthropic/claude-sonnet-5"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh", "max"]
@@ -16,3 +20,7 @@ cache_write = 1.7
 
 [limit]
 output = 64000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index db0f2598960..5365c3ba5d9 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 40.00% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
index 8a90f294b0b..cddc13d6092 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
 base_model = "deepseek/deepseek-v4-flash"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
index 99b8adf9f26..91cf6381a19 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro-0813"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
index 12150561530..8887fd0de20 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index 844b3b874ea..c1f8bb03ac7 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.6 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/gemini-2.5-flash.toml b/providers/cheaperinference/models/gemini-2.5-flash.toml
index e7b20d51473..8bd00409f33 100644
--- a/providers/cheaperinference/models/gemini-2.5-flash.toml
+++ b/providers/cheaperinference/models/gemini-2.5-flash.toml
@@ -1,7 +1,9 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
 base_model = "google/gemini-2.5-flash"
-
 reasoning = false
 
 [cost]
diff --git a/providers/cheaperinference/models/gemini-3-5-flash.toml b/providers/cheaperinference/models/gemini-3-5-flash.toml
index 384c3717312..3b55bce9f04 100644
--- a/providers/cheaperinference/models/gemini-3-5-flash.toml
+++ b/providers/cheaperinference/models/gemini-3-5-flash.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.5/$9.0 per 1M).
 base_model = "google/gemini-3.5-flash"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["minimal", "low", "medium", "high"]
diff --git a/providers/cheaperinference/models/gemini-3-flash-preview.toml b/providers/cheaperinference/models/gemini-3-flash-preview.toml
index bf8da3e4f71..ec9817ed551 100644
--- a/providers/cheaperinference/models/gemini-3-flash-preview.toml
+++ b/providers/cheaperinference/models/gemini-3-flash-preview.toml
@@ -2,7 +2,8 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.5/$3.0 per 1M).
 base_model = "google/gemini-3-flash-preview"
 
-attachment = false
+[interleaved]
+field = "reasoning_details"
 
 [[reasoning_options]]
 type = "effort"
diff --git a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
index 6e19627093c..949205853eb 100644
--- a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
+++ b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
@@ -1,7 +1,9 @@
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$1.5 per 1M).
 base_model = "google/gemini-3.1-flash-lite"
-
 reasoning = false
 
 [cost]
diff --git a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
index f42ddb44228..7f49ec856b9 100644
--- a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
+++ b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
@@ -2,7 +2,8 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$12.0 per 1M).
 base_model = "google/gemini-3.1-pro-preview"
 
-attachment = false
+[interleaved]
+field = "reasoning_content"
 
 [[reasoning_options]]
 type = "effort"
diff --git a/providers/cheaperinference/models/gemini-3.7-flash.toml b/providers/cheaperinference/models/gemini-3.7-flash.toml
index e975ecacf23..181618fc486 100644
--- a/providers/cheaperinference/models/gemini-3.7-flash.toml
+++ b/providers/cheaperinference/models/gemini-3.7-flash.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$3.75 per 1M).
 base_model = "google/gemini-3.7-flash"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high"]
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
index 5d55a21220f..5477594f4fa 100644
--- a/providers/cheaperinference/models/glm-4.5-air.toml
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
 base_model = "zhipuai/glm-4.5-air"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
index 45aad806e7e..2d2ffd63d08 100644
--- a/providers/cheaperinference/models/glm-4.5.toml
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.5"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
index e447239e1ba..dc17181d10a 100644
--- a/providers/cheaperinference/models/glm-4.6.toml
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.6"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
index d3180d0bfb0..5859c78be7d 100644
--- a/providers/cheaperinference/models/glm-4.7.toml
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -2,6 +2,7 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.7"
+interleaved = true
 
 [[reasoning_options]]
 type = "toggle"
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
index 51f3ddfd839..bdff367adba 100644
--- a/providers/cheaperinference/models/glm-5.1.toml
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.1"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/glm-5.2.toml b/providers/cheaperinference/models/glm-5.2.toml
index ad3609d303d..4fe65ced60f 100644
--- a/providers/cheaperinference/models/glm-5.2.toml
+++ b/providers/cheaperinference/models/glm-5.2.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.8/$2.55 per 1M).
 base_model = "zhipuai/glm-5.2"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["high", "max"]
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
index b382619c492..80a1abe2d8d 100644
--- a/providers/cheaperinference/models/glm-5.3-flash.toml
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 46.29% off the model maker's list price ($0.15/$0.5 per 1M).
 base_model = "zhipuai/glm-5.3-flash"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
index 9c3d855d110..5e891686d20 100644
--- a/providers/cheaperinference/models/glm-5.3.toml
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.3"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "high", "max"]
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
index 12d4e21ed6e..5f6348cc800 100644
--- a/providers/cheaperinference/models/glm-5.toml
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
 base_model = "zhipuai/glm-5"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
index 3e34216873f..17838bae4a8 100644
--- a/providers/cheaperinference/models/gpt-5-mini.toml
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -1,7 +1,8 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$2.0 per 1M).
 base_model = "openai/gpt-5-mini"
-
 attachment = false
 
 [[reasoning_options]]
@@ -13,3 +14,7 @@ input = 0.201421
 output = 1.611374
 cache_read = 0.020142
 cache_write = 0.201421
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-nano.toml b/providers/cheaperinference/models/gpt-5-nano.toml
index 179c285f58f..9f2173851ea 100644
--- a/providers/cheaperinference/models/gpt-5-nano.toml
+++ b/providers/cheaperinference/models/gpt-5-nano.toml
@@ -1,7 +1,8 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.05/$0.4 per 1M).
 base_model = "openai/gpt-5-nano"
-
 attachment = false
 
 [[reasoning_options]]
@@ -13,3 +14,7 @@ input = 0.040284
 output = 0.322274
 cache_read = 0.004028
 cache_write = 0.040284
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.2-codex.toml b/providers/cheaperinference/models/gpt-5.2-codex.toml
index b071ab80686..bf7d024498d 100644
--- a/providers/cheaperinference/models/gpt-5.2-codex.toml
+++ b/providers/cheaperinference/models/gpt-5.2-codex.toml
@@ -1,9 +1,13 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.75/$14.0 per 1M).
 base_model = "openai/gpt-5.2-codex"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high", "xhigh"]
@@ -17,3 +21,7 @@ cache_write = 1.225
 [limit]
 context = 256000
 output = 65536
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4-mini.toml b/providers/cheaperinference/models/gpt-5.4-mini.toml
index 79f01368462..7688dee4c08 100644
--- a/providers/cheaperinference/models/gpt-5.4-mini.toml
+++ b/providers/cheaperinference/models/gpt-5.4-mini.toml
@@ -1,7 +1,8 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$4.5 per 1M).
 base_model = "openai/gpt-5.4-mini"
-
 attachment = false
 
 [[reasoning_options]]
@@ -13,3 +14,7 @@ input = 0.525
 output = 3.15
 cache_read = 0.0525
 cache_write = 0.525
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4.toml b/providers/cheaperinference/models/gpt-5.4.toml
index 01ace763f69..9fc30ed617e 100644
--- a/providers/cheaperinference/models/gpt-5.4.toml
+++ b/providers/cheaperinference/models/gpt-5.4.toml
@@ -1,7 +1,8 @@
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$15.0 per 1M).
 base_model = "openai/gpt-5.4"
-
 attachment = false
 
 [[reasoning_options]]
@@ -17,3 +18,7 @@ cache_write = 1.75
 [limit]
 context = 1000000
 output = 131072
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.5-pro.toml b/providers/cheaperinference/models/gpt-5.5-pro.toml
index ca07f180ee8..36e41df6ae8 100644
--- a/providers/cheaperinference/models/gpt-5.5-pro.toml
+++ b/providers/cheaperinference/models/gpt-5.5-pro.toml
@@ -1,10 +1,13 @@
+# /v1/chat/completions returns 400 "Pro reasoning mode requires the /v1/responses endpoint", and /v1/responses
+# answers, so this route is responses-shaped. There effort medium and high are accepted, low returns 400 and
+# xhigh fails upstream (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($30.0/$180.0 per 1M).
 base_model = "openai/gpt-5.5-pro"
 
 [[reasoning_options]]
 type = "effort"
-values = ["medium", "high", "xhigh"]
+values = ["medium", "high"]
 
 [cost]
 input = 21.0
@@ -14,3 +17,6 @@ cache_write = 21.0
 
 [limit]
 context = 1000000
+
+[provider]
+shape = "responses"
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
index eca6ca5f23f..bf6b35ffce7 100644
--- a/providers/cheaperinference/models/gpt-oss-120b.toml
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
 base_model = "openai/gpt-oss-120b"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high"]
diff --git a/providers/cheaperinference/models/grok-4.5.toml b/providers/cheaperinference/models/grok-4.5.toml
index 96ae58cd9c6..de2d61ee690 100644
--- a/providers/cheaperinference/models/grok-4.5.toml
+++ b/providers/cheaperinference/models/grok-4.5.toml
@@ -2,6 +2,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$6.0 per 1M).
 base_model = "xai/grok-4.5"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high"]
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index d346a7b616e..a9df71c310b 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
index 0417f4da7c7..3762381d917 100644
--- a/providers/cheaperinference/models/minimax-m2.7.toml
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "minimax/MiniMax-M2.7"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/qwen-3-8-27b.toml b/providers/cheaperinference/models/qwen-3-8-27b.toml
index 9426e7b0c8b..c4a80857876 100644
--- a/providers/cheaperinference/models/qwen-3-8-27b.toml
+++ b/providers/cheaperinference/models/qwen-3-8-27b.toml
@@ -1,10 +1,17 @@
+# Toggle: reasoning.enabled = true|false
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.214/$2.55 per 1M).
 base_model = "alibaba/qwen3.8-27b"
 
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
 [[reasoning_options]]
 type = "effort"
-values = ["none", "low", "medium", "xhigh"]
+values = ["low", "medium", "xhigh"]
 
 [cost]
 input = 0.1498
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index 4bbd1e6d7a8..eeed2e8b0ab 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -1,12 +1,22 @@
+# Reasoning is always on and takes no caller control: reasoning_content comes back on every request, including
+# with reasoning.enabled = false and with reasoning.effort = "none" (6 of 6 probed 2026-09-16).
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
-
 attachment = false
 reasoning_options = []
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 1.75
 output = 5.25
 cache_read = 0.2125
 cache_write = 2.125
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
index c8606655fef..df2958b7d1c 100644
--- a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -1,10 +1,14 @@
 # Toggle: reasoning.enabled = true|false
+# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
+# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
 base_model = "alibaba/qwen3.5-35b-a3b"
-
 attachment = false
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
@@ -16,3 +20,7 @@ cache_write = 0.21875
 
 [limit]
 context = 256000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
index bb117766f63..46311cbc317 100644
--- a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.1/$1.0 per 1M).
 base_model = "alibaba/qwen3.6-35b-a3b"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
index cfd7181f23d..ad50ee84c12 100644
--- a/providers/cheaperinference/models/qwen3.6-27b.toml
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -3,6 +3,9 @@
 # Discounted marketplace rate, 55.24% off the model maker's list price ($0.6/$3.6 per 1M).
 base_model = "alibaba/qwen3.6-27b"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "toggle"
 

From 88d9d3ef4d5a287ac06c8c442fca2525a31562aa Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 02:06:08 +0400
Subject: [PATCH 12/17] Record probed image and reasoning behaviour on five
 CheaperInference routes

claude-fable-5, claude-fable-5.1 and claude-opus-5-fast advertise capabilities.vision
true, but an image_url part is rejected on every attempt (502 or 400, "all available
provider routes failed"), exactly like the other Anthropic routes on this host, so they
get attachment = false and a text-only input modality. All three do return the reasoning
side channel in reasoning_content, so interleaved stays.

gemini-3.6-flash reports usage.completion_tokens_details.reasoning_tokens but the message
carries only content on 8 of 8 probes, so no side channel is declared for it.

muse-spark-1.2 returns its reasoning in reasoning_details on 3 of 3 untruncated probes
(runs cut short by max_tokens carry none), so interleaved.field is set accordingly.
---
 providers/cheaperinference/models/claude-fable-5.1.toml   | 7 +++++++
 providers/cheaperinference/models/claude-fable-5.toml     | 7 +++++++
 providers/cheaperinference/models/claude-opus-5-fast.toml | 7 +++++++
 providers/cheaperinference/models/gemini-3.6-flash.toml   | 3 +++
 providers/cheaperinference/models/muse-spark-1.2.toml     | 5 +++++
 5 files changed, 29 insertions(+)

diff --git a/providers/cheaperinference/models/claude-fable-5.1.toml b/providers/cheaperinference/models/claude-fable-5.1.toml
index 123d05632d5..483a37c3ec1 100644
--- a/providers/cheaperinference/models/claude-fable-5.1.toml
+++ b/providers/cheaperinference/models/claude-fable-5.1.toml
@@ -1,6 +1,9 @@
+# No image input on this host: an image_url part is rejected (400, every upstream route failing) on 2 of 2
+# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5-1"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -14,3 +17,7 @@ input = 7.0
 output = 35.0
 cache_read = 0.175
 cache_write = 8.5
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-fable-5.toml b/providers/cheaperinference/models/claude-fable-5.toml
index f492eabf1d8..24f1524ae91 100644
--- a/providers/cheaperinference/models/claude-fable-5.toml
+++ b/providers/cheaperinference/models/claude-fable-5.toml
@@ -1,6 +1,9 @@
+# No image input on this host: an image_url part is rejected (502, every upstream route failing) on 2 of 2
+# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -14,3 +17,7 @@ input = 7.0
 output = 35.0
 cache_read = 0.7
 cache_write = 8.5
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-5-fast.toml b/providers/cheaperinference/models/claude-opus-5-fast.toml
index 9b2cbe78529..51f9394731e 100644
--- a/providers/cheaperinference/models/claude-opus-5-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-5-fast.toml
@@ -1,6 +1,9 @@
+# No image input on this host: an image_url part is rejected (400, every upstream route failing) on 2 of 2
+# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-opus-5"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -14,3 +17,7 @@ input = 7.0
 output = 35.0
 cache_read = 0.7
 cache_write = 8.5
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gemini-3.6-flash.toml b/providers/cheaperinference/models/gemini-3.6-flash.toml
index f4b04ede6d6..eb6ea77ffee 100644
--- a/providers/cheaperinference/models/gemini-3.6-flash.toml
+++ b/providers/cheaperinference/models/gemini-3.6-flash.toml
@@ -1,3 +1,6 @@
+# Reasoning is billed and counted but never returned: 8 of 8 probes reported
+# usage.completion_tokens_details.reasoning_tokens (235-385) while the message carried only content,
+# so there is no side channel to declare here (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 15.00% off the model maker's list price ($0.75/$3.75 per 1M).
 base_model = "google/gemini-3.6-flash"
diff --git a/providers/cheaperinference/models/muse-spark-1.2.toml b/providers/cheaperinference/models/muse-spark-1.2.toml
index f4adcb6ef8a..d6096f1d054 100644
--- a/providers/cheaperinference/models/muse-spark-1.2.toml
+++ b/providers/cheaperinference/models/muse-spark-1.2.toml
@@ -1,7 +1,12 @@
+# Reasoning is returned in a reasoning_details field on the message: present on 3 of 3 untruncated probes
+# (runs cut short by max_tokens carry none), so the side channel is declared below (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
 # Discounted marketplace rate, 15.00% off the model maker's list price ($1.25/$4.25 per 1M).
 base_model = "meta/muse-spark-1.2"
 
+[interleaved]
+field = "reasoning_details"
+
 [[reasoning_options]]
 type = "effort"
 values = ["minimal", "low", "medium", "high", "xhigh"]

From ef9663cd0e59379511bea9c3f3aa3439df0af3e3 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 02:07:59 +0400
Subject: [PATCH 13/17] Report an unresearched CheaperInference reasoner
 instead of failing the sync

A model that reasons on this host but exposes no reasoning controls in the catalog
still needs hand-authored reasoning_options. Throwing a plain Error aborted the whole
run, so a single new model could silence the hourly sync for every other model.

MissingReasoningOptionsError is the shape the runner already understands: it keeps the
local file, records the ID and carries on, so the gap is reported without stopping the
run.
---
 packages/core/src/sync/providers/cheaperinference.ts | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/packages/core/src/sync/providers/cheaperinference.ts b/packages/core/src/sync/providers/cheaperinference.ts
index 2f369d34b3a..4ede4f2d823 100644
--- a/packages/core/src/sync/providers/cheaperinference.ts
+++ b/packages/core/src/sync/providers/cheaperinference.ts
@@ -1,6 +1,7 @@
 import { z } from "zod";
 
 import type { ExistingModel, SyncedFullModel, SyncedModel, SyncProvider } from "../index.js";
+import { MissingReasoningOptionsError } from "../missing-reasoning-options.js";
 import { factorBaseModel } from "./openrouter.js";
 
 const API_ENDPOINT = "https://api.cheaperinference.com/v1/models";
@@ -133,8 +134,11 @@ export function buildCheaperInferenceModel(
   existing: ExistingModel,
 ): SyncedModel {
   if (existing.reasoning !== false && existing.reasoning_options === undefined) {
-    throw new Error(
-      `CheaperInference model ${model.id} requires hand-authored reasoning_options; the catalog exposes no reasoning controls`,
+    // The runner keeps the local file and reports the ID instead of failing the
+    // whole sync, so one unresearched model cannot stop the hourly run.
+    throw new MissingReasoningOptionsError(
+      model.id,
+      "reasons on this host but the catalog exposes no reasoning controls, so reasoning_options must be hand-authored",
     );
   }
 

From b1a546342980c07dba47a29cedfc2029a6645dc6 Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 02:09:57 +0400
Subject: [PATCH 14/17] Cover the reported-not-fatal path for an unresearched
 CheaperInference reasoner

The old assertion only matched the message text, so a plain Error would have passed it.
It now asserts the error type and the model ID it carries.

A second test drives syncProvider over a temporary models directory holding one
researched and one unresearched reasoner, and checks what the run actually does with
them: the unresearched file is left byte-for-byte alone, nothing is deleted, the other
model still syncs, and the ID is handed to the missing-model issue opener.
---
 packages/core/test/cheaperinference.test.ts | 57 +++++++++++++++++++--
 1 file changed, 53 insertions(+), 4 deletions(-)

diff --git a/packages/core/test/cheaperinference.test.ts b/packages/core/test/cheaperinference.test.ts
index 0f687650dfe..f2265ea9e32 100644
--- a/packages/core/test/cheaperinference.test.ts
+++ b/packages/core/test/cheaperinference.test.ts
@@ -1,5 +1,10 @@
-import { expect, test } from "bun:test";
+import { expect, spyOn, test } from "bun:test";
+import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import path from "node:path";
 
+import { syncProvider } from "../src/sync/index.js";
+import * as missingIssues from "../src/sync/missing-issues.js";
+import { MissingReasoningOptionsError } from "../src/sync/missing-reasoning-options.js";
 import {
   buildCheaperInferenceModel,
   CheaperInferenceResponse,
@@ -98,12 +103,56 @@ test("takes the gateway's limits when it publishes them and inherits when it doe
 });
 
 test("refuses to sync a reasoning model that has no authored controls", () => {
-  expect(() =>
+  // MissingReasoningOptionsError, not a bare Error: the runner catches this one
+  // to skip the single ID and keep the local file, instead of failing the run.
+  let thrown: unknown;
+  try {
     buildCheaperInferenceModel(sourceModel() as never, {
       ...existing,
       reasoning_options: undefined,
-    } as never),
-  ).toThrow(/reasoning_options/);
+    } as never);
+  } catch (error) {
+    thrown = error;
+  }
+  expect(thrown).toBeInstanceOf(MissingReasoningOptionsError);
+  expect((thrown as MissingReasoningOptionsError).modelId).toBe("claude-sonnet-5");
+  expect((thrown as Error).message).toContain("reasoning_options");
+});
+
+test("an unresearched reasoner is reported and its file kept, and the rest of the run continues", async () => {
+  const dir = await mkdtemp(path.join(import.meta.dirname, "../../../providers/.cheaperinference-sync-"));
+  const modelsDir = path.join(dir, "models");
+  const authored = '# authored by hand\nbase_model = "anthropic/claude-sonnet-5"\n';
+  const researched =
+    '# authored by hand\nbase_model = "anthropic/claude-opus-5"\n\n[[reasoning_options]]\ntype = "effort"\nvalues = ["low", "high"]\n';
+  await mkdir(modelsDir, { recursive: true });
+  await writeFile(path.join(modelsDir, "claude-sonnet-5.toml"), authored);
+  await writeFile(path.join(modelsDir, "claude-opus-5.toml"), researched);
+  const issues = spyOn(missingIssues, "openMissingModelIssues").mockResolvedValue([]);
+  try {
+    const result = await syncProvider(
+      {
+        ...cheaperinference,
+        modelsDir,
+        async fetchModels() {
+          return {
+            object: "list",
+            data: [sourceModel(), sourceModel({ id: "claude-opus-5" })],
+            pricing_version: "sha256:test",
+            pricing_checked_at: "2026-09-16T08:00:58.592Z",
+          } as never;
+        },
+      },
+      { openIssues: true },
+    );
+    expect(result.deleted).toBe(0);
+    expect(await readFile(path.join(modelsDir, "claude-sonnet-5.toml"), "utf8")).toBe(authored);
+    expect(issues.mock.calls[0]?.[1]).toEqual(["claude-sonnet-5"]);
+    expect(issues.mock.calls[0]?.[2]?.reasons?.["claude-sonnet-5"]).toContain("reasoning_options");
+  } finally {
+    issues.mockRestore();
+    await rm(dir, { recursive: true, force: true });
+  }
 });
 
 test("only syncs token-priced text routes", () => {

From 3969c1fe1bf592fbd91e9c91efa2fd46e064483b Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 11:51:05 +0400
Subject: [PATCH 15/17] Re-probe image input with a valid image and refresh the
 catalogue rates

The earlier image-input probe used a 1x1 PNG. Upstreams treat that as junk: routes that handle a
normal image returned 400 or 502 "all available provider routes failed", and some that do accept images
answered "White" for a red pixel. Every attachment and modalities value derived from that probe was
wrong, in both directions.

Re-probed with a 32x32 solid-colour PNG as a data: URI, four colours per model, max_tokens high enough
that a reasoning model still answers, and a text-only control on every route that errored, so "no image
input" is separated from "route down". A route counts as having image input only when it names the
colour; several accept the request and then answer NOIMAGE or guess.

Image input works on 36 of the 59 chat routes reachable on /v1/chat/completions, plus gpt-5.5-pro on the
/v1/responses layer it requires. Each file states what was measured. Two notes:

- deepseek-v4-flash-0731 reads images here although its base model is text-only (10 of 10 probes), so it
  carries an explicit attachment and modalities override rather than inheriting.
- A remote https image URL is rejected by every route tested; only data: URIs work.

Rates re-read at the same time (pricing_version sha256:6d2520b9, 2026-09-17T07:00:28Z): deepseek-v4.1-flash
doubled, glm-5.3-flash and muse-spark-1.2 fell, gpt-5.4-nano rose. The other rates are unchanged.
---
 .../models/claude-fable-5.1.toml              | 11 +++--------
 .../models/claude-fable-5.toml                | 11 +++--------
 .../models/claude-haiku-4.5.toml              |  4 +++-
 .../models/claude-opus-4-8-fast.toml          |  4 +++-
 .../models/claude-opus-4.5.toml               |  9 +++------
 .../models/claude-opus-4.6.toml               |  9 +++------
 .../models/claude-opus-4.7.toml               |  9 +++------
 .../models/claude-opus-4.8.toml               | 11 +++--------
 .../models/claude-opus-5-fast.toml            | 11 +++--------
 .../models/claude-opus-5.toml                 | 11 +++--------
 .../models/claude-sonnet-4.5.toml             |  9 +++------
 .../models/claude-sonnet-4.6.toml             |  9 +++------
 .../models/claude-sonnet-5.toml               |  6 +++---
 .../models/deepseek-v4-flash-0731.toml        | 10 +++++++++-
 .../models/deepseek-v4-flash.toml             |  9 ++++++++-
 .../models/deepseek-v4-pro-0813.toml          |  9 ++++++++-
 .../models/deepseek-v4-pro.toml               |  9 ++++++++-
 .../models/deepseek-v4.1-flash.toml           | 14 ++++++++------
 .../models/gemini-2.5-flash.toml              |  4 +++-
 .../models/gemini-3-5-flash.toml              |  4 +++-
 .../models/gemini-3-flash-preview.toml        |  4 +++-
 .../models/gemini-3.1-flash-lite.toml         |  4 +++-
 .../models/gemini-3.1-pro-preview.toml        |  4 +++-
 .../models/gemini-3.6-flash.toml              |  4 +++-
 .../models/gemini-3.7-flash.toml              |  4 +++-
 .../cheaperinference/models/glm-4.5-air.toml  |  9 ++++++++-
 .../cheaperinference/models/glm-4.5.toml      |  9 ++++++++-
 .../cheaperinference/models/glm-4.6.toml      |  9 ++++++++-
 .../cheaperinference/models/glm-4.7.toml      |  9 ++++++++-
 .../cheaperinference/models/glm-5.1.toml      |  9 ++++++++-
 .../cheaperinference/models/glm-5.2.toml      |  9 ++++++++-
 .../models/glm-5.3-flash.toml                 | 17 ++++++++++++-----
 .../cheaperinference/models/glm-5.3.toml      |  9 ++++++++-
 providers/cheaperinference/models/glm-5.toml  |  9 ++++++++-
 .../cheaperinference/models/gpt-4.1-nano.toml |  9 ++++++++-
 .../cheaperinference/models/gpt-5-mini.toml   |  6 +++---
 .../cheaperinference/models/gpt-5-nano.toml   |  6 +++---
 .../models/gpt-5.2-codex.toml                 | 11 +++--------
 .../cheaperinference/models/gpt-5.4-mini.toml | 11 +++--------
 .../cheaperinference/models/gpt-5.4-nano.toml | 19 +++++++++++++------
 .../cheaperinference/models/gpt-5.4.toml      | 11 +++--------
 .../cheaperinference/models/gpt-5.5-pro.toml  |  5 ++++-
 .../cheaperinference/models/gpt-5.5.toml      |  4 +++-
 .../cheaperinference/models/gpt-5.6-luna.toml |  4 +++-
 .../cheaperinference/models/gpt-5.6-sol.toml  |  4 +++-
 .../models/gpt-5.6-terra.toml                 |  4 +++-
 .../cheaperinference/models/gpt-6-astra.toml  |  4 +++-
 .../cheaperinference/models/gpt-oss-120b.toml |  9 ++++++++-
 .../cheaperinference/models/grok-4.5.toml     |  4 +++-
 .../cheaperinference/models/kimi-k3.toml      |  4 +++-
 .../cheaperinference/models/minimax-m2.7.toml |  9 ++++++++-
 .../models/muse-spark-1.2.toml                | 19 +++++++++++++------
 .../cheaperinference/models/qwen-3-8-27b.toml |  4 +++-
 .../cheaperinference/models/qwen-3-8-max.toml |  9 +++------
 .../models/qwen3-5-35b-a3b.toml               |  9 +++------
 .../models/qwen3-6-35b-a3b.toml               |  9 ++++++++-
 .../cheaperinference/models/qwen3.6-27b.toml  |  4 +++-
 57 files changed, 287 insertions(+), 173 deletions(-)

diff --git a/providers/cheaperinference/models/claude-fable-5.1.toml b/providers/cheaperinference/models/claude-fable-5.1.toml
index 483a37c3ec1..a28cc5472b2 100644
--- a/providers/cheaperinference/models/claude-fable-5.1.toml
+++ b/providers/cheaperinference/models/claude-fable-5.1.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, every upstream route failing) on 2 of 2
-# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5-1"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 7.0
 output = 35.0
 cache_read = 0.175
 cache_write = 8.5
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-fable-5.toml b/providers/cheaperinference/models/claude-fable-5.toml
index 24f1524ae91..fb04e3bf7be 100644
--- a/providers/cheaperinference/models/claude-fable-5.toml
+++ b/providers/cheaperinference/models/claude-fable-5.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (502, every upstream route failing) on 2 of 2
-# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-fable-5"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 7.0
 output = 35.0
 cache_read = 0.7
 cache_write = 8.5
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-haiku-4.5.toml b/providers/cheaperinference/models/claude-haiku-4.5.toml
index 216949b787c..8066af50629 100644
--- a/providers/cheaperinference/models/claude-haiku-4.5.toml
+++ b/providers/cheaperinference/models/claude-haiku-4.5.toml
@@ -1,7 +1,9 @@
+# Image input verified on this host: 2 of 4 colour probes exact, the others "lime" for green and
+# "magenta" for purple (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
 # thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
 # reports capabilities.reasoning false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.0/$5.0 per 1M).
 base_model = "anthropic/claude-haiku-4-5"
 reasoning = false
diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
index cddfc43a14d..ab26ea4ea9d 100644
--- a/providers/cheaperinference/models/claude-opus-4-8-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -1,7 +1,9 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 400 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
 attachment = false
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
index 2942f72ca24..32401910e9c 100644
--- a/providers/cheaperinference/models/claude-opus-4.5.toml
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -1,12 +1,13 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-5"
-attachment = false
 reasoning = false
 
 [cost]
@@ -18,7 +19,3 @@ cache_write = 4.25
 [limit]
 context = 198000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
index 519be6b1f9f..e153d392fb7 100644
--- a/providers/cheaperinference/models/claude-opus-4.6.toml
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -1,12 +1,13 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-6"
-attachment = false
 reasoning = false
 
 [cost]
@@ -14,7 +15,3 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.7.toml b/providers/cheaperinference/models/claude-opus-4.7.toml
index a820ea31c7c..0e0d4c53396 100644
--- a/providers/cheaperinference/models/claude-opus-4.7.toml
+++ b/providers/cheaperinference/models/claude-opus-4.7.toml
@@ -1,12 +1,13 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
 # thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
 # reports capabilities.reasoning false (probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-7"
-attachment = false
 reasoning = false
 
 [cost]
@@ -14,7 +15,3 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.8.toml b/providers/cheaperinference/models/claude-opus-4.8.toml
index fe8debe0a3a..e1597404760 100644
--- a/providers/cheaperinference/models/claude-opus-4.8.toml
+++ b/providers/cheaperinference/models/claude-opus-4.8.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-5-fast.toml b/providers/cheaperinference/models/claude-opus-5-fast.toml
index 51f9394731e..df7310cc5cd 100644
--- a/providers/cheaperinference/models/claude-opus-5-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-5-fast.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, every upstream route failing) on 2 of 2
-# requests, even though the catalog reports capabilities.vision true (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "anthropic/claude-opus-5"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 7.0
 output = 35.0
 cache_read = 0.7
 cache_write = 8.5
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-5.toml b/providers/cheaperinference/models/claude-opus-5.toml
index c83539c9b82..3ae53009f66 100644
--- a/providers/cheaperinference/models/claude-opus-5.toml
+++ b/providers/cheaperinference/models/claude-opus-5.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part was rejected on two of three requests and the catalog reports
-# capabilities.vision false, so image input is not dependable here (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-5"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 3.5
 output = 17.5
 cache_read = 0.35
 cache_write = 4.25
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
index 64d697cf707..e12bfca20b0 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -1,12 +1,13 @@
+# Image input verified on this host: 3 of 4 colour probes exact and the fourth called "lime" for
+# green (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-5"
-attachment = false
 reasoning = false
 
 [cost]
@@ -17,7 +18,3 @@ cache_write = 2.55
 
 [limit]
 context = 198000
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
index cde1f93590d..ac4c1db25cc 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.6.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -1,12 +1,13 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-6"
-attachment = false
 reasoning = false
 
 [cost]
@@ -14,7 +15,3 @@ input = 2.1
 output = 10.5
 cache_read = 0.21
 cache_write = 2.55
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 1adde594198..0e6a9151fb2 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -1,6 +1,6 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "anthropic/claude-sonnet-5"
 attachment = false
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index 5365c3ba5d9..a18954ac650 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -1,7 +1,11 @@
+# Image input works on this host although the base model is text-only: a 32x32 solid-colour PNG as a
+# data: URI was named correctly on 10 of 10 probes across four colours, so attachment and the input
+# modalities are overridden here rather than inherited (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 40.00% off the model maker's list price ($0.076/$0.153 per 1M).
 base_model = "deepseek/deepseek-v4-flash-0731"
+attachment = true
 
 [interleaved]
 field = "reasoning_content"
@@ -22,3 +26,7 @@ cache_write = 0.0456
 [limit]
 context = 1048576
 output = 65536
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
index cddc13d6092..35e6da563ba 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -1,7 +1,10 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE for green and purple on 4 of 4 probes, with only the blue guess landing (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
 base_model = "deepseek/deepseek-v4-flash"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,3 +24,7 @@ cache_write = 0.071388
 
 [limit]
 output = 128000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
index 91cf6381a19..f242c4483d4 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -1,7 +1,10 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE for green and purple on 4 of 4 probes, with only the blue guess landing (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro-0813"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,3 +24,7 @@ cache_write = 0.462
 
 [limit]
 context = 1048576
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
index 8887fd0de20..d4faa048d97 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -1,7 +1,10 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 6 of 6 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,3 +24,7 @@ cache_write = 0.462
 
 [limit]
 output = 128000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
index c1f8bb03ac7..0483ac30e78 100644
--- a/providers/cheaperinference/models/deepseek-v4.1-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -1,6 +1,8 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
-# Discounted marketplace rate, 30.00% off the model maker's list price ($0.15/$0.6 per 1M).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "deepseek/deepseek-v4.1-flash"
 
 [interleaved]
@@ -14,10 +16,10 @@ type = "effort"
 values = ["low", "high", "max"]
 
 [cost]
-input = 0.105
-output = 0.42
-cache_read = 0.0021
-cache_write = 0.105
+input = 0.21
+output = 0.84
+cache_read = 0.0042
+cache_write = 0.21
 
 [limit]
 context = 1048576
diff --git a/providers/cheaperinference/models/gemini-2.5-flash.toml b/providers/cheaperinference/models/gemini-2.5-flash.toml
index 8bd00409f33..82054bb36ca 100644
--- a/providers/cheaperinference/models/gemini-2.5-flash.toml
+++ b/providers/cheaperinference/models/gemini-2.5-flash.toml
@@ -1,7 +1,9 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
 # thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
 # reports capabilities.reasoning false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
 base_model = "google/gemini-2.5-flash"
 reasoning = false
diff --git a/providers/cheaperinference/models/gemini-3-5-flash.toml b/providers/cheaperinference/models/gemini-3-5-flash.toml
index 3b55bce9f04..974352eeb29 100644
--- a/providers/cheaperinference/models/gemini-3-5-flash.toml
+++ b/providers/cheaperinference/models/gemini-3-5-flash.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.5/$9.0 per 1M).
 base_model = "google/gemini-3.5-flash"
 
diff --git a/providers/cheaperinference/models/gemini-3-flash-preview.toml b/providers/cheaperinference/models/gemini-3-flash-preview.toml
index ec9817ed551..d892bebf1c7 100644
--- a/providers/cheaperinference/models/gemini-3-flash-preview.toml
+++ b/providers/cheaperinference/models/gemini-3-flash-preview.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.5/$3.0 per 1M).
 base_model = "google/gemini-3-flash-preview"
 
diff --git a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
index 949205853eb..8600ee38911 100644
--- a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
+++ b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
@@ -1,7 +1,9 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
 # thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
 # reports capabilities.reasoning false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$1.5 per 1M).
 base_model = "google/gemini-3.1-flash-lite"
 reasoning = false
diff --git a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
index 7f49ec856b9..f312c3a785d 100644
--- a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
+++ b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$12.0 per 1M).
 base_model = "google/gemini-3.1-pro-preview"
 
diff --git a/providers/cheaperinference/models/gemini-3.6-flash.toml b/providers/cheaperinference/models/gemini-3.6-flash.toml
index eb6ea77ffee..f4eee3e05ea 100644
--- a/providers/cheaperinference/models/gemini-3.6-flash.toml
+++ b/providers/cheaperinference/models/gemini-3.6-flash.toml
@@ -1,7 +1,9 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Reasoning is billed and counted but never returned: 8 of 8 probes reported
 # usage.completion_tokens_details.reasoning_tokens (235-385) while the message carried only content,
 # so there is no side channel to declare here (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 15.00% off the model maker's list price ($0.75/$3.75 per 1M).
 base_model = "google/gemini-3.6-flash"
 
diff --git a/providers/cheaperinference/models/gemini-3.7-flash.toml b/providers/cheaperinference/models/gemini-3.7-flash.toml
index 181618fc486..10fbb304fff 100644
--- a/providers/cheaperinference/models/gemini-3.7-flash.toml
+++ b/providers/cheaperinference/models/gemini-3.7-flash.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$3.75 per 1M).
 base_model = "google/gemini-3.7-flash"
 
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
index 5477594f4fa..1d0cbbb65c5 100644
--- a/providers/cheaperinference/models/glm-4.5-air.toml
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: thinking.type = enabled|disabled
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
 base_model = "zhipuai/glm-4.5-air"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -18,3 +21,7 @@ cache_write = 0.104739
 [limit]
 context = 128000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
index 2d2ffd63d08..8f319c90fc7 100644
--- a/providers/cheaperinference/models/glm-4.5.toml
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.5"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -18,3 +21,7 @@ cache_write = 0.33
 [limit]
 context = 128000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
index dc17181d10a..8bf78d02972 100644
--- a/providers/cheaperinference/models/glm-4.6.toml
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.6"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -18,3 +21,7 @@ cache_write = 0.33
 [limit]
 context = 198000
 output = 16384
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
index 5859c78be7d..cbe82d50a7b 100644
--- a/providers/cheaperinference/models/glm-4.7.toml
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.7"
+attachment = false
 interleaved = true
 
 [[reasoning_options]]
@@ -16,3 +19,7 @@ cache_write = 0.322274
 [limit]
 context = 128000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
index bdff367adba..d1ea86a04f5 100644
--- a/providers/cheaperinference/models/glm-5.1.toml
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.1"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,3 +20,7 @@ cache_write = 0.77
 
 [limit]
 output = 24000
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.2.toml b/providers/cheaperinference/models/glm-5.2.toml
index 4fe65ced60f..5f5676c8afd 100644
--- a/providers/cheaperinference/models/glm-5.2.toml
+++ b/providers/cheaperinference/models/glm-5.2.toml
@@ -1,6 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.8/$2.55 per 1M).
 base_model = "zhipuai/glm-5.2"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,3 +20,7 @@ cache_write = 0.44
 
 [limit]
 context = 1048576
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
index 80a1abe2d8d..059cb44aefd 100644
--- a/providers/cheaperinference/models/glm-5.3-flash.toml
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -1,6 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
-# Discounted marketplace rate, 46.29% off the model maker's list price ($0.15/$0.5 per 1M).
+# No dependable image input on this host: 10 probes gave 3 colours right, "red" for green and one
+# NOIMAGE, so image input cannot be relied on here (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 51.66% off the model maker's list price ($0.15/$0.5 per 1M).
 base_model = "zhipuai/glm-5.3-flash"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -10,10 +13,14 @@ type = "effort"
 values = ["low", "high", "max"]
 
 [cost]
-input = 0.080568
-output = 0.268535
+input = 0.072511
+output = 0.241706
 cache_read = 0.01275
-cache_write = 0.080568
+cache_write = 0.072511
 
 [limit]
 context = 1048576
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
index 5e891686d20..dcae99d11e5 100644
--- a/providers/cheaperinference/models/glm-5.3.toml
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -1,6 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.3"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -14,3 +17,7 @@ input = 0.77
 output = 2.42
 cache_read = 0.119
 cache_write = 0.77
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
index 5f6348cc800..69e207a261e 100644
--- a/providers/cheaperinference/models/glm-5.toml
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 6 of 6 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
 base_model = "zhipuai/glm-5"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -18,3 +21,7 @@ cache_write = 0.483412
 [limit]
 context = 198000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-4.1-nano.toml b/providers/cheaperinference/models/gpt-4.1-nano.toml
index b4ec41095b4..a2ea5e1cf56 100644
--- a/providers/cheaperinference/models/gpt-4.1-nano.toml
+++ b/providers/cheaperinference/models/gpt-4.1-nano.toml
@@ -1,9 +1,16 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 15.00% off the model maker's list price ($0.1/$0.4 per 1M).
 base_model = "openai/gpt-4.1-nano"
+attachment = false
 
 [cost]
 input = 0.085
 output = 0.34
 cache_read = 0.008499
 cache_write = 0.085
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
index 17838bae4a8..f579ef8ba06 100644
--- a/providers/cheaperinference/models/gpt-5-mini.toml
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -1,6 +1,6 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the colour is guessed - 0 of 6 probes
+# right (orange, maroon, olive, magenta, cyan, turquoise) (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$2.0 per 1M).
 base_model = "openai/gpt-5-mini"
 attachment = false
diff --git a/providers/cheaperinference/models/gpt-5-nano.toml b/providers/cheaperinference/models/gpt-5-nano.toml
index 9f2173851ea..abf78ba77fc 100644
--- a/providers/cheaperinference/models/gpt-5-nano.toml
+++ b/providers/cheaperinference/models/gpt-5-nano.toml
@@ -1,6 +1,6 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the colour is guessed - 1 of 6 probes
+# right, the rest wrong or empty (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 19.43% off the model maker's list price ($0.05/$0.4 per 1M).
 base_model = "openai/gpt-5-nano"
 attachment = false
diff --git a/providers/cheaperinference/models/gpt-5.2-codex.toml b/providers/cheaperinference/models/gpt-5.2-codex.toml
index bf7d024498d..4755e4c25b5 100644
--- a/providers/cheaperinference/models/gpt-5.2-codex.toml
+++ b/providers/cheaperinference/models/gpt-5.2-codex.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($1.75/$14.0 per 1M).
 base_model = "openai/gpt-5.2-codex"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 1.225
 [limit]
 context = 256000
 output = 65536
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4-mini.toml b/providers/cheaperinference/models/gpt-5.4-mini.toml
index 7688dee4c08..e75899b3609 100644
--- a/providers/cheaperinference/models/gpt-5.4-mini.toml
+++ b/providers/cheaperinference/models/gpt-5.4-mini.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$4.5 per 1M).
 base_model = "openai/gpt-5.4-mini"
-attachment = false
 
 [[reasoning_options]]
 type = "effort"
@@ -14,7 +13,3 @@ input = 0.525
 output = 3.15
 cache_read = 0.0525
 cache_write = 0.525
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4-nano.toml b/providers/cheaperinference/models/gpt-5.4-nano.toml
index 2287a5d60ab..c54cab6f2a4 100644
--- a/providers/cheaperinference/models/gpt-5.4-nano.toml
+++ b/providers/cheaperinference/models/gpt-5.4-nano.toml
@@ -1,13 +1,20 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
-# Discounted marketplace rate, 30.00% off the model maker's list price ($0.2/$1.25 per 1M).
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# "black" twice and NOIMAGE twice across 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.2/$1.25 per 1M).
 base_model = "openai/gpt-5.4-nano"
+attachment = false
 
 [[reasoning_options]]
 type = "effort"
 values = ["none", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 0.14
-output = 0.875
-cache_read = 0.014
-cache_write = 0.14
+input = 0.161137
+output = 1.007109
+cache_read = 0.016114
+cache_write = 0.161137
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4.toml b/providers/cheaperinference/models/gpt-5.4.toml
index 9fc30ed617e..c8045ce6fb1 100644
--- a/providers/cheaperinference/models/gpt-5.4.toml
+++ b/providers/cheaperinference/models/gpt-5.4.toml
@@ -1,9 +1,8 @@
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: 3 of 4 colour probes exact and the fourth called "orange" for
+# yellow (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$15.0 per 1M).
 base_model = "openai/gpt-5.4"
-attachment = false
 
 [[reasoning_options]]
 type = "effort"
@@ -18,7 +17,3 @@ cache_write = 1.75
 [limit]
 context = 1000000
 output = 131072
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.5-pro.toml b/providers/cheaperinference/models/gpt-5.5-pro.toml
index 36e41df6ae8..b26a4104979 100644
--- a/providers/cheaperinference/models/gpt-5.5-pro.toml
+++ b/providers/cheaperinference/models/gpt-5.5-pro.toml
@@ -1,7 +1,10 @@
+# Image input verified on this host through the /v1/responses layer this route requires: a 32x32
+# solid-colour PNG as a data: URI was named correctly on 4 of 5 attempts, the fifth a transient 502
+# (probed 2026-09-17).
 # /v1/chat/completions returns 400 "Pro reasoning mode requires the /v1/responses endpoint", and /v1/responses
 # answers, so this route is responses-shaped. There effort medium and high are accepted, low returns 400 and
 # xhigh fails upstream (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($30.0/$180.0 per 1M).
 base_model = "openai/gpt-5.5-pro"
 
diff --git a/providers/cheaperinference/models/gpt-5.5.toml b/providers/cheaperinference/models/gpt-5.5.toml
index bb844b8055c..4aa5dcf0a60 100644
--- a/providers/cheaperinference/models/gpt-5.5.toml
+++ b/providers/cheaperinference/models/gpt-5.5.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$30.0 per 1M).
 base_model = "openai/gpt-5.5"
 
diff --git a/providers/cheaperinference/models/gpt-5.6-luna.toml b/providers/cheaperinference/models/gpt-5.6-luna.toml
index ea2f4c8af95..f97d8800b7d 100644
--- a/providers/cheaperinference/models/gpt-5.6-luna.toml
+++ b/providers/cheaperinference/models/gpt-5.6-luna.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.2/$1.2 per 1M).
 base_model = "openai/gpt-5.6-luna"
 
diff --git a/providers/cheaperinference/models/gpt-5.6-sol.toml b/providers/cheaperinference/models/gpt-5.6-sol.toml
index f650d64b021..bf26e1ee568 100644
--- a/providers/cheaperinference/models/gpt-5.6-sol.toml
+++ b/providers/cheaperinference/models/gpt-5.6-sol.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 50.00% off the model maker's list price ($2.0/$10.0 per 1M).
 base_model = "openai/gpt-5.6-sol"
 
diff --git a/providers/cheaperinference/models/gpt-5.6-terra.toml b/providers/cheaperinference/models/gpt-5.6-terra.toml
index d6cc346b6dc..fd948ab09fc 100644
--- a/providers/cheaperinference/models/gpt-5.6-terra.toml
+++ b/providers/cheaperinference/models/gpt-5.6-terra.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($2.0/$12.0 per 1M).
 base_model = "openai/gpt-5.6-terra"
 
diff --git a/providers/cheaperinference/models/gpt-6-astra.toml b/providers/cheaperinference/models/gpt-6-astra.toml
index cfa5462e77f..6f8f65529fe 100644
--- a/providers/cheaperinference/models/gpt-6-astra.toml
+++ b/providers/cheaperinference/models/gpt-6-astra.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
 base_model = "openai/gpt-6-astra"
 
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
index bf6b35ffce7..600e30c9023 100644
--- a/providers/cheaperinference/models/gpt-oss-120b.toml
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -1,6 +1,9 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
 base_model = "openai/gpt-oss-120b"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -14,3 +17,7 @@ input = 0.04
 output = 0.2
 cache_read = 0.0085
 cache_write = 0.04
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/grok-4.5.toml b/providers/cheaperinference/models/grok-4.5.toml
index de2d61ee690..6ca9c1491a8 100644
--- a/providers/cheaperinference/models/grok-4.5.toml
+++ b/providers/cheaperinference/models/grok-4.5.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$6.0 per 1M).
 base_model = "xai/grok-4.5"
 
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
index a9df71c310b..9d11d32f6ca 100644
--- a/providers/cheaperinference/models/kimi-k3.toml
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -1,5 +1,7 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "moonshotai/kimi-k3"
 
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
index 3762381d917..5586b586923 100644
--- a/providers/cheaperinference/models/minimax-m2.7.toml
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -1,7 +1,10 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "minimax/MiniMax-M2.7"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -18,3 +21,7 @@ cache_write = 0.255
 [limit]
 context = 198000
 output = 32768
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/muse-spark-1.2.toml b/providers/cheaperinference/models/muse-spark-1.2.toml
index d6096f1d054..698d3ee4435 100644
--- a/providers/cheaperinference/models/muse-spark-1.2.toml
+++ b/providers/cheaperinference/models/muse-spark-1.2.toml
@@ -1,8 +1,11 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE or an empty message on 6 of 6 colour probes (probed 2026-09-17).
 # Reasoning is returned in a reasoning_details field on the message: present on 3 of 3 untruncated probes
 # (runs cut short by max_tokens carry none), so the side channel is declared below (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
-# Discounted marketplace rate, 15.00% off the model maker's list price ($1.25/$4.25 per 1M).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.25/$4.25 per 1M).
 base_model = "meta/muse-spark-1.2"
+attachment = false
 
 [interleaved]
 field = "reasoning_details"
@@ -12,10 +15,14 @@ type = "effort"
 values = ["minimal", "low", "medium", "high", "xhigh"]
 
 [cost]
-input = 1.0625
-output = 3.6125
-cache_read = 0.10625
-cache_write = 1.0625
+input = 0.875
+output = 2.975
+cache_read = 0.105
+cache_write = 0.875
 
 [limit]
 output = 943718
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/qwen-3-8-27b.toml b/providers/cheaperinference/models/qwen-3-8-27b.toml
index c4a80857876..c3879cd149a 100644
--- a/providers/cheaperinference/models/qwen-3-8-27b.toml
+++ b/providers/cheaperinference/models/qwen-3-8-27b.toml
@@ -1,5 +1,7 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.214/$2.55 per 1M).
 base_model = "alibaba/qwen3.8-27b"
 
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index eeed2e8b0ab..2a71a9cf53a 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -1,11 +1,12 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Reasoning is always on and takes no caller control: reasoning_content comes back on every request, including
 # with reasoning.enabled = false and with reasoning.effort = "none" (6 of 6 probed 2026-09-16).
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
-attachment = false
 reasoning_options = []
 
 [interleaved]
@@ -16,7 +17,3 @@ input = 1.75
 output = 5.25
 cache_read = 0.2125
 cache_write = 2.125
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
index df2958b7d1c..328edea0be9 100644
--- a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -1,10 +1,11 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
 # No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
 # catalog reports capabilities.vision false (probed 2026-09-16).
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
 base_model = "alibaba/qwen3.5-35b-a3b"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -20,7 +21,3 @@ cache_write = 0.21875
 
 [limit]
 context = 256000
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
index 46311cbc317..aece80e1925 100644
--- a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -1,7 +1,10 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 400 on 6 of 6 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.1/$1.0 per 1M).
 base_model = "alibaba/qwen3.6-35b-a3b"
+attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,3 +20,7 @@ cache_write = 0.07
 
 [limit]
 output = 262144
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
index ad50ee84c12..89568257521 100644
--- a/providers/cheaperinference/models/qwen3.6-27b.toml
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -1,5 +1,7 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 55.24% off the model maker's list price ($0.6/$3.6 per 1M).
 base_model = "alibaba/qwen3.6-27b"
 

From e19c6f176bf9e73d4a85ba3344ecc0d85a7a51ed Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 11:52:32 +0400
Subject: [PATCH 16/17] Re-probe image input on the prefixed
 gemini-3.5-flash-lite route

Same 32x32 data: URI probe as the sibling routes: 4 of 4 colours named correctly, so this route keeps
image input. Rates re-read at the same catalogue version.
---
 .../cheaperinference/models/google/gemini-3.5-flash-lite.toml | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
index 3cdced4c1ab..8c0e1534430 100644
--- a/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
+++ b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
@@ -1,4 +1,6 @@
-# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:1099de72…, checked 2026-09-16T08:00:58Z)
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
 base_model = "google/gemini-3.5-flash-lite"
 

From 83d1fa5bcfa72b412f4f3ac73c7b4d5b3778af2f Mon Sep 17 00:00:00 2001
From: aiapienthusiast <327629069+aiapienthusiast@users.noreply.github.com>
Date: Thu, 17 Sep 2026 16:04:00 +0400
Subject: [PATCH 17/17] Keep the Cheaper Inference files override-only and drop
 the stale probe notes

Override-only rule: 13 text-only entries restated attachment = false and a
[modalities] block identical to the lab entry, and 10 more restated
modalities.output. Those lines are gone; base_model merges deeply, so the
generated output is byte-identical (checked with bun run validate before and
after).

Headers: the 2026-09-16 no-image note is removed from the 8 files where the
2026-09-17 probe supersedes it - 7 that now verify image input, and
claude-opus-4-8-fast, where it repeated the newer no-image note.

bun run validate exits 0; bun test packages/core is 305 pass / 3 fail, the same
3 failures as on the branch without this change.
---
 providers/cheaperinference/models/claude-opus-4-8-fast.toml  | 3 ---
 providers/cheaperinference/models/claude-opus-4.5.toml       | 2 --
 providers/cheaperinference/models/claude-opus-4.6.toml       | 2 --
 providers/cheaperinference/models/claude-opus-4.7.toml       | 2 --
 providers/cheaperinference/models/claude-sonnet-4.5.toml     | 2 --
 providers/cheaperinference/models/claude-sonnet-4.6.toml     | 2 --
 providers/cheaperinference/models/claude-sonnet-5.toml       | 1 -
 .../cheaperinference/models/deepseek-v4-flash-0731.toml      | 1 -
 providers/cheaperinference/models/deepseek-v4-flash.toml     | 5 -----
 providers/cheaperinference/models/deepseek-v4-pro-0813.toml  | 5 -----
 providers/cheaperinference/models/deepseek-v4-pro.toml       | 5 -----
 providers/cheaperinference/models/glm-4.5-air.toml           | 5 -----
 providers/cheaperinference/models/glm-4.5.toml               | 5 -----
 providers/cheaperinference/models/glm-4.6.toml               | 5 -----
 providers/cheaperinference/models/glm-4.7.toml               | 5 -----
 providers/cheaperinference/models/glm-5.1.toml               | 5 -----
 providers/cheaperinference/models/glm-5.2.toml               | 5 -----
 providers/cheaperinference/models/glm-5.3-flash.toml         | 1 -
 providers/cheaperinference/models/glm-5.3.toml               | 5 -----
 providers/cheaperinference/models/glm-5.toml                 | 5 -----
 providers/cheaperinference/models/gpt-4.1-nano.toml          | 1 -
 providers/cheaperinference/models/gpt-5-mini.toml            | 1 -
 providers/cheaperinference/models/gpt-5-nano.toml            | 1 -
 providers/cheaperinference/models/gpt-5.4-nano.toml          | 1 -
 providers/cheaperinference/models/gpt-oss-120b.toml          | 5 -----
 providers/cheaperinference/models/minimax-m2.7.toml          | 5 -----
 providers/cheaperinference/models/muse-spark-1.2.toml        | 1 -
 providers/cheaperinference/models/qwen-3-8-max.toml          | 2 --
 providers/cheaperinference/models/qwen3-5-35b-a3b.toml       | 2 --
 providers/cheaperinference/models/qwen3-6-35b-a3b.toml       | 1 -
 30 files changed, 91 deletions(-)

diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
index ab26ea4ea9d..6ce5f09b443 100644
--- a/providers/cheaperinference/models/claude-opus-4-8-fast.toml
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -1,8 +1,6 @@
 # No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
 # 400 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
 base_model = "anthropic/claude-opus-4-8"
@@ -26,4 +24,3 @@ cache_write = 10.2
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
index 32401910e9c..4586bb922b2 100644
--- a/providers/cheaperinference/models/claude-opus-4.5.toml
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -3,8 +3,6 @@
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-5"
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
index e153d392fb7..07f499ad23d 100644
--- a/providers/cheaperinference/models/claude-opus-4.6.toml
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -3,8 +3,6 @@
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-6"
diff --git a/providers/cheaperinference/models/claude-opus-4.7.toml b/providers/cheaperinference/models/claude-opus-4.7.toml
index 0e0d4c53396..4af692312f0 100644
--- a/providers/cheaperinference/models/claude-opus-4.7.toml
+++ b/providers/cheaperinference/models/claude-opus-4.7.toml
@@ -3,8 +3,6 @@
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
 # thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
 # reports capabilities.reasoning false (probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
 base_model = "anthropic/claude-opus-4-7"
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
index e12bfca20b0..28babdc5771 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -3,8 +3,6 @@
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-5"
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
index ac4c1db25cc..d04b32f738a 100644
--- a/providers/cheaperinference/models/claude-sonnet-4.6.toml
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -3,8 +3,6 @@
 # No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
 # 200 with no reasoning_content and no reasoning token count, though the catalog reports
 # capabilities.reasoning true (probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
 base_model = "anthropic/claude-sonnet-4-6"
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
index 0e6a9151fb2..e55c244600a 100644
--- a/providers/cheaperinference/models/claude-sonnet-5.toml
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -23,4 +23,3 @@ output = 64000
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
index a18954ac650..a9301d2e38a 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -29,4 +29,3 @@ output = 65536
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
index 35e6da563ba..2fd837482f3 100644
--- a/providers/cheaperinference/models/deepseek-v4-flash.toml
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
 base_model = "deepseek/deepseek-v4-flash"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -24,7 +23,3 @@ cache_write = 0.071388
 
 [limit]
 output = 128000
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
index f242c4483d4..bba5653d0a2 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro-0813"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -24,7 +23,3 @@ cache_write = 0.462
 
 [limit]
 context = 1048576
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
index d4faa048d97..6c999fa5252 100644
--- a/providers/cheaperinference/models/deepseek-v4-pro.toml
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
 base_model = "deepseek/deepseek-v4-pro"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -24,7 +23,3 @@ cache_write = 0.462
 
 [limit]
 output = 128000
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
index 1d0cbbb65c5..a94d1e11ce8 100644
--- a/providers/cheaperinference/models/glm-4.5-air.toml
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
 base_model = "zhipuai/glm-4.5-air"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 0.104739
 [limit]
 context = 128000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
index 8f319c90fc7..92935999ec3 100644
--- a/providers/cheaperinference/models/glm-4.5.toml
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.5"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 0.33
 [limit]
 context = 128000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
index 8bf78d02972..6d000fa4889 100644
--- a/providers/cheaperinference/models/glm-4.6.toml
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.6"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 0.33
 [limit]
 context = 198000
 output = 16384
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
index cbe82d50a7b..aa25e6b9ae7 100644
--- a/providers/cheaperinference/models/glm-4.7.toml
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
 base_model = "zhipuai/glm-4.7"
-attachment = false
 interleaved = true
 
 [[reasoning_options]]
@@ -19,7 +18,3 @@ cache_write = 0.322274
 [limit]
 context = 128000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
index d1ea86a04f5..2de36652906 100644
--- a/providers/cheaperinference/models/glm-5.1.toml
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.1"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -20,7 +19,3 @@ cache_write = 0.77
 
 [limit]
 output = 24000
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.2.toml b/providers/cheaperinference/models/glm-5.2.toml
index 5f5676c8afd..18aa327e9ed 100644
--- a/providers/cheaperinference/models/glm-5.2.toml
+++ b/providers/cheaperinference/models/glm-5.2.toml
@@ -3,7 +3,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($0.8/$2.55 per 1M).
 base_model = "zhipuai/glm-5.2"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -20,7 +19,3 @@ cache_write = 0.44
 
 [limit]
 context = 1048576
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
index 059cb44aefd..2b0a819a45b 100644
--- a/providers/cheaperinference/models/glm-5.3-flash.toml
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -23,4 +23,3 @@ context = 1048576
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
index dcae99d11e5..22332be6ffc 100644
--- a/providers/cheaperinference/models/glm-5.3.toml
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -3,7 +3,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
 base_model = "zhipuai/glm-5.3"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 0.77
 output = 2.42
 cache_read = 0.119
 cache_write = 0.77
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
index 69e207a261e..d013708de60 100644
--- a/providers/cheaperinference/models/glm-5.toml
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
 base_model = "zhipuai/glm-5"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 0.483412
 [limit]
 context = 198000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-4.1-nano.toml b/providers/cheaperinference/models/gpt-4.1-nano.toml
index a2ea5e1cf56..eea038d8246 100644
--- a/providers/cheaperinference/models/gpt-4.1-nano.toml
+++ b/providers/cheaperinference/models/gpt-4.1-nano.toml
@@ -13,4 +13,3 @@ cache_write = 0.085
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
index f579ef8ba06..579cbddba16 100644
--- a/providers/cheaperinference/models/gpt-5-mini.toml
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -17,4 +17,3 @@ cache_write = 0.201421
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-nano.toml b/providers/cheaperinference/models/gpt-5-nano.toml
index abf78ba77fc..dc2627003cb 100644
--- a/providers/cheaperinference/models/gpt-5-nano.toml
+++ b/providers/cheaperinference/models/gpt-5-nano.toml
@@ -17,4 +17,3 @@ cache_write = 0.040284
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4-nano.toml b/providers/cheaperinference/models/gpt-5.4-nano.toml
index c54cab6f2a4..35272eca19d 100644
--- a/providers/cheaperinference/models/gpt-5.4-nano.toml
+++ b/providers/cheaperinference/models/gpt-5.4-nano.toml
@@ -17,4 +17,3 @@ cache_write = 0.161137
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
index 600e30c9023..1ee0cbfe40c 100644
--- a/providers/cheaperinference/models/gpt-oss-120b.toml
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -3,7 +3,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
 base_model = "openai/gpt-oss-120b"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -17,7 +16,3 @@ input = 0.04
 output = 0.2
 cache_read = 0.0085
 cache_write = 0.04
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
index 5586b586923..0654c123111 100644
--- a/providers/cheaperinference/models/minimax-m2.7.toml
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -4,7 +4,6 @@
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
 base_model = "minimax/MiniMax-M2.7"
-attachment = false
 
 [interleaved]
 field = "reasoning_content"
@@ -21,7 +20,3 @@ cache_write = 0.255
 [limit]
 context = 198000
 output = 32768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/muse-spark-1.2.toml b/providers/cheaperinference/models/muse-spark-1.2.toml
index 698d3ee4435..e1d025ea6a4 100644
--- a/providers/cheaperinference/models/muse-spark-1.2.toml
+++ b/providers/cheaperinference/models/muse-spark-1.2.toml
@@ -25,4 +25,3 @@ output = 943718
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
index 2a71a9cf53a..dc324f00d5f 100644
--- a/providers/cheaperinference/models/qwen-3-8-max.toml
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -2,8 +2,6 @@
 # 4 of 4 colour probes (probed 2026-09-17).
 # Reasoning is always on and takes no caller control: reasoning_content comes back on every request, including
 # with reasoning.enabled = false and with reasoning.effort = "none" (6 of 6 probed 2026-09-16).
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
 base_model = "alibaba/qwen3.8-max"
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
index 328edea0be9..2e609965597 100644
--- a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -1,8 +1,6 @@
 # Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
 # 4 of 4 colour probes (probed 2026-09-17).
 # Toggle: reasoning.enabled = true|false
-# No image input on this host: an image_url part is rejected (400, or every upstream route failing) and the
-# catalog reports capabilities.vision false (probed 2026-09-16).
 # Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
 # Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
 base_model = "alibaba/qwen3.5-35b-a3b"
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
index aece80e1925..7a047967e25 100644
--- a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -23,4 +23,3 @@ output = 262144
 
 [modalities]
 input = ["text"]
-output = ["text"]