From 84a8e1b7b3347b23b7b156b1210b5aad1311c777 Mon Sep 17 00:00:00 2001
From: K-3-LT <76082878+K-3-LT@users.noreply.github.com>
Date: Tue, 15 Sep 2026 14:35:26 +0800
Subject: [PATCH 1/2] feat: update IteraCompute model catalog

---
 .../deepseek/deepseek-v4-flash-0731.toml      | 16 ++++++++++
 .../models/deepseek/deepseek-v4-pro-0813.toml | 16 ++++++++++
 .../iteracompute/ornith-1.5-35b-a3b.toml      | 31 -------------------
 .../models/iteracompute/qwen3.8-27b.toml      | 24 --------------
 .../models/minimax/minimax-m3.toml            | 19 ++++++++++++
 .../models/moonshotai/kimi-k3.toml            | 20 ++++++++++++
 .../models/ornith-ai/ornith-1.5-35b-a3b.toml  | 22 +++++++++++++
 .../models/qwen/qwen3.8-2.4t-a95b.toml        | 19 ++++++++++++
 .../iteracompute/models/qwen/qwen3.8-27b.toml | 19 ++++++++++++
 .../models/z-ai/glm-5.3-flash.toml            | 20 ++++++++++++
 .../iteracompute/models/z-ai/glm-5.3.toml     | 17 ++++++++++
 11 files changed, 168 insertions(+), 55 deletions(-)
 create mode 100644 providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
 create mode 100644 providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
 delete mode 100644 providers/iteracompute/models/iteracompute/ornith-1.5-35b-a3b.toml
 delete mode 100644 providers/iteracompute/models/iteracompute/qwen3.8-27b.toml
 create mode 100644 providers/iteracompute/models/minimax/minimax-m3.toml
 create mode 100644 providers/iteracompute/models/moonshotai/kimi-k3.toml
 create mode 100644 providers/iteracompute/models/ornith-ai/ornith-1.5-35b-a3b.toml
 create mode 100644 providers/iteracompute/models/qwen/qwen3.8-2.4t-a95b.toml
 create mode 100644 providers/iteracompute/models/qwen/qwen3.8-27b.toml
 create mode 100644 providers/iteracompute/models/z-ai/glm-5.3-flash.toml
 create mode 100644 providers/iteracompute/models/z-ai/glm-5.3.toml

diff --git a/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml b/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
new file mode 100644
index 00000000000..e90d275db10
--- /dev/null
+++ b/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
@@ -0,0 +1,16 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Reasoning is mandatory and the public catalog exposes no caller-selectable reasoning control.
+# Prices are in USD per million tokens.
+base_model = "deepseek/deepseek-v4-flash-0731"
+reasoning_options = []
+
+[cost]
+input = 0.35
+output = 1.30
+cache_read = 0.03
+
+[limit]
+context = 970_000
+output = 393_216
diff --git a/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml b/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
new file mode 100644
index 00000000000..a7681a06e75
--- /dev/null
+++ b/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
@@ -0,0 +1,16 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Reasoning is mandatory and the public catalog exposes no caller-selectable reasoning control.
+# Prices are in USD per million tokens.
+base_model = "deepseek/deepseek-v4-pro-0813"
+reasoning_options = []
+
+[cost]
+input = 1.10
+output = 3.50
+cache_read = 0.11
+
+[limit]
+context = 1_048_576
+output = 393_216
diff --git a/providers/iteracompute/models/iteracompute/ornith-1.5-35b-a3b.toml b/providers/iteracompute/models/iteracompute/ornith-1.5-35b-a3b.toml
deleted file mode 100644
index 59d4eb1b5ea..00000000000
--- a/providers/iteracompute/models/iteracompute/ornith-1.5-35b-a3b.toml
+++ /dev/null
@@ -1,31 +0,0 @@
-# Sources (accessed 2026-08-28):
-# https://api.iteracompute.com/v1/models (model ID, availability, limits, and pricing)
-# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
-# https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B (upstream capabilities)
-#
-# IteraCompute serves the language-only NVFP4 build with a 327,680-token
-# combined window, up to 262,144 input tokens, and up to 65,536 output tokens.
-# Reasoning toggle wire path: reasoning_effort = "none" disables thinking;
-# any accepted non-none effort enables it. Responses stream reasoning through
-# the reasoning_content delta field.
-# Prices are in USD per million tokens.
-base_model = "deepreinforce/ornith-1.5-35b-a3b"
-attachment = false
-
-[interleaved]
-field = "reasoning_content"
-
-[[reasoning_options]]
-type = "toggle"
-
-[cost]
-input = 0.30
-output = 3.00
-
-[limit]
-context = 327_680
-input = 262_144
-output = 65_536
-
-[modalities]
-input = ["text"]
diff --git a/providers/iteracompute/models/iteracompute/qwen3.8-27b.toml b/providers/iteracompute/models/iteracompute/qwen3.8-27b.toml
deleted file mode 100644
index 8eb4170a5f0..00000000000
--- a/providers/iteracompute/models/iteracompute/qwen3.8-27b.toml
+++ /dev/null
@@ -1,24 +0,0 @@
-# Sources (accessed 2026-09-01):
-# https://api.iteracompute.com/v1/models (model ID, availability, and input/output pricing)
-# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
-# https://huggingface.co/gittensor-model-hub/Qwen3.8-27B-NVFP4-RTX5090 (served checkpoint and image input)
-#
-# IteraCompute serves Qwen3.8 27B through its OpenAI-compatible API.
-# The NVFP4 deployment accepts text and image input and extends the combined
-# serving window to 327,680 tokens with up to 262,144 input and 65,536 output tokens.
-# Prices are in USD per million tokens.
-# Effort: reasoning_effort = none|low|medium|xhigh (xhigh is the default).
-base_model = "alibaba/qwen3.8-27b"
-reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "xhigh"] }]
-
-[cost]
-input = 0.30
-output = 2.50
-
-[limit]
-context = 327_680
-input = 262_144
-output = 65_536
-
-[modalities]
-input = ["text", "image"]
diff --git a/providers/iteracompute/models/minimax/minimax-m3.toml b/providers/iteracompute/models/minimax/minimax-m3.toml
new file mode 100644
index 00000000000..10f663ec30d
--- /dev/null
+++ b/providers/iteracompute/models/minimax/minimax-m3.toml
@@ -0,0 +1,19 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Reasoning is mandatory and the public catalog exposes no caller-selectable reasoning control.
+# Prices are in USD per million tokens.
+base_model = "minimax/MiniMax-M3"
+reasoning_options = []
+
+[cost]
+input = 0.40
+output = 1.60
+cache_read = 0.08
+
+[limit]
+context = 1_048_576
+output = 524_288
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/iteracompute/models/moonshotai/kimi-k3.toml b/providers/iteracompute/models/moonshotai/kimi-k3.toml
new file mode 100644
index 00000000000..b6055d6489f
--- /dev/null
+++ b/providers/iteracompute/models/moonshotai/kimi-k3.toml
@@ -0,0 +1,20 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Reasoning is mandatory and the public catalog exposes no caller-selectable reasoning control.
+# Prices are in USD per million tokens.
+base_model = "moonshotai/kimi-k3"
+structured_output = false
+reasoning_options = []
+
+[cost]
+input = 3.10
+output = 15.50
+cache_read = 0.31
+
+[limit]
+context = 1_048_576
+output = 999_999
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/iteracompute/models/ornith-ai/ornith-1.5-35b-a3b.toml b/providers/iteracompute/models/ornith-ai/ornith-1.5-35b-a3b.toml
new file mode 100644
index 00000000000..e0ee4a079e5
--- /dev/null
+++ b/providers/iteracompute/models/ornith-ai/ornith-1.5-35b-a3b.toml
@@ -0,0 +1,22 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (canonical model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Toggle wire path: reasoning_effort = "none" disables thinking; responses expose reasoning_content.
+# Prices are in USD per million tokens.
+base_model = "deepreinforce/ornith-1.5-35b-a3b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.30
+output = 3.00
+cache_read = 0.03
+
+[limit]
+context = 327_680
+input = 262_144
+output = 65_536
diff --git a/providers/iteracompute/models/qwen/qwen3.8-2.4t-a95b.toml b/providers/iteracompute/models/qwen/qwen3.8-2.4t-a95b.toml
new file mode 100644
index 00000000000..1499c45a580
--- /dev/null
+++ b/providers/iteracompute/models/qwen/qwen3.8-2.4t-a95b.toml
@@ -0,0 +1,19 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Prices are in USD per million tokens.
+base_model = "alibaba/qwen3.8-2.4t-a95b"
+attachment = true
+reasoning_options = [{ type = "toggle" }]
+
+[cost]
+input = 1.95
+output = 5.95
+cache_read = 0.20
+
+[limit]
+context = 970_000
+output = 131_072
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/iteracompute/models/qwen/qwen3.8-27b.toml b/providers/iteracompute/models/qwen/qwen3.8-27b.toml
new file mode 100644
index 00000000000..bb7bc16c39d
--- /dev/null
+++ b/providers/iteracompute/models/qwen/qwen3.8-27b.toml
@@ -0,0 +1,19 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (canonical model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Prices are in USD per million tokens.
+base_model = "alibaba/qwen3.8-27b"
+reasoning_options = [{ type = "effort", values = ["none", "low", "medium", "xhigh"] }]
+
+[cost]
+input = 0.30
+output = 2.50
+cache_read = 0.03
+
+[limit]
+context = 327_680
+input = 262_144
+output = 65_536
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/iteracompute/models/z-ai/glm-5.3-flash.toml b/providers/iteracompute/models/z-ai/glm-5.3-flash.toml
new file mode 100644
index 00000000000..2f3eea506f0
--- /dev/null
+++ b/providers/iteracompute/models/z-ai/glm-5.3-flash.toml
@@ -0,0 +1,20 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, pricing, and capacity)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Effort wire path: reasoning.effort = low|high|max. Reasoning is mandatory.
+# Prices are in USD per million tokens.
+base_model = "zhipuai/glm-5.3-flash"
+structured_output = false
+reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }]
+
+[cost]
+input = 0.12
+output = 0.48
+cache_read = 0.02
+
+[limit]
+context = 1_048_576
+output = 131_072
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/iteracompute/models/z-ai/glm-5.3.toml b/providers/iteracompute/models/z-ai/glm-5.3.toml
new file mode 100644
index 00000000000..04a97272db5
--- /dev/null
+++ b/providers/iteracompute/models/z-ai/glm-5.3.toml
@@ -0,0 +1,17 @@
+# Sources (accessed 2026-09-15):
+# https://api.iteracompute.com/v1/models (model ID, availability, capabilities, limits, and pricing)
+# https://iteracompute.com/docs.html (OpenAI-compatible API documentation)
+# Reasoning is mandatory and the public catalog exposes no caller-selectable reasoning control.
+# Prices are in USD per million tokens.
+base_model = "zhipuai/glm-5.3"
+structured_output = false
+reasoning_options = []
+
+[cost]
+input = 1.50
+output = 6.20
+cache_read = 0.23
+
+[limit]
+context = 1_048_576
+output = 131_072

From ffc5b282e835e77f3b87b924436fa5d0d317eba2 Mon Sep 17 00:00:00 2001
From: Tao <Z1197756683@outlook.com>
Date: Wed, 16 Sep 2026 16:50:12 +0800
Subject: [PATCH 2/2] fix: sync IteraCompute prices with production catalog

---
 .../models/deepseek/deepseek-v4-flash-0731.toml             | 6 +++---
 .../iteracompute/models/deepseek/deepseek-v4-pro-0813.toml  | 2 +-
 providers/iteracompute/models/minimax/minimax-m3.toml       | 4 ++--
 providers/iteracompute/models/moonshotai/kimi-k3.toml       | 6 +++---
 providers/iteracompute/models/z-ai/glm-5.3-flash.toml       | 6 +++---
 providers/iteracompute/models/z-ai/glm-5.3.toml             | 6 +++---
 6 files changed, 15 insertions(+), 15 deletions(-)

diff --git a/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml b/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
index e90d275db10..de987dfd5fe 100644
--- a/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
+++ b/providers/iteracompute/models/deepseek/deepseek-v4-flash-0731.toml
@@ -7,9 +7,9 @@ base_model = "deepseek/deepseek-v4-flash-0731"
 reasoning_options = []
 
 [cost]
-input = 0.35
-output = 1.30
-cache_read = 0.03
+input = 0.34
+output = 1.05
+cache_read = 0.035
 
 [limit]
 context = 970_000
diff --git a/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml b/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
index a7681a06e75..0199acb7930 100644
--- a/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
+++ b/providers/iteracompute/models/deepseek/deepseek-v4-pro-0813.toml
@@ -8,7 +8,7 @@ reasoning_options = []
 
 [cost]
 input = 1.10
-output = 3.50
+output = 3.30
 cache_read = 0.11
 
 [limit]
diff --git a/providers/iteracompute/models/minimax/minimax-m3.toml b/providers/iteracompute/models/minimax/minimax-m3.toml
index 10f663ec30d..f9d9908e3f6 100644
--- a/providers/iteracompute/models/minimax/minimax-m3.toml
+++ b/providers/iteracompute/models/minimax/minimax-m3.toml
@@ -7,8 +7,8 @@ base_model = "minimax/MiniMax-M3"
 reasoning_options = []
 
 [cost]
-input = 0.40
-output = 1.60
+input = 0.29
+output = 1.20
 cache_read = 0.08
 
 [limit]
diff --git a/providers/iteracompute/models/moonshotai/kimi-k3.toml b/providers/iteracompute/models/moonshotai/kimi-k3.toml
index b6055d6489f..4e4af1d1240 100644
--- a/providers/iteracompute/models/moonshotai/kimi-k3.toml
+++ b/providers/iteracompute/models/moonshotai/kimi-k3.toml
@@ -8,9 +8,9 @@ structured_output = false
 reasoning_options = []
 
 [cost]
-input = 3.10
-output = 15.50
-cache_read = 0.31
+input = 3.00
+output = 14.90
+cache_read = 0.29
 
 [limit]
 context = 1_048_576
diff --git a/providers/iteracompute/models/z-ai/glm-5.3-flash.toml b/providers/iteracompute/models/z-ai/glm-5.3-flash.toml
index 2f3eea506f0..19503c99421 100644
--- a/providers/iteracompute/models/z-ai/glm-5.3-flash.toml
+++ b/providers/iteracompute/models/z-ai/glm-5.3-flash.toml
@@ -8,9 +8,9 @@ structured_output = false
 reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }]
 
 [cost]
-input = 0.12
-output = 0.48
-cache_read = 0.02
+input = 0.14
+output = 0.49
+cache_read = 0.03
 
 [limit]
 context = 1_048_576
diff --git a/providers/iteracompute/models/z-ai/glm-5.3.toml b/providers/iteracompute/models/z-ai/glm-5.3.toml
index 04a97272db5..2ba1f3cca79 100644
--- a/providers/iteracompute/models/z-ai/glm-5.3.toml
+++ b/providers/iteracompute/models/z-ai/glm-5.3.toml
@@ -8,9 +8,9 @@ structured_output = false
 reasoning_options = []
 
 [cost]
-input = 1.50
-output = 6.20
-cache_read = 0.23
+input = 1.20
+output = 3.50
+cache_read = 0.26
 
 [limit]
 context = 1_048_576