From b4d1868f2c07df6b06db34eee6972a123b97c27a Mon Sep 17 00:00:00 2001
From: Oliver Mee <102673257+oliver-mee@users.noreply.github.com>
Date: Tue, 15 Sep 2026 15:43:32 +0800
Subject: [PATCH 1/2] feat(alibaba-token-plan): add deepseek-v4.1-flash
 (international gateway)

---
 .../models/deepseek-v4.1-flash.toml           | 27 +++++++++++++++++++
 1 file changed, 27 insertions(+)
 create mode 100644 providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml

diff --git a/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml b/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml
new file mode 100644
index 00000000000..57cf02bed6c
--- /dev/null
+++ b/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml
@@ -0,0 +1,27 @@
+# Reasoning HTTP format (accessed 2026-09-15):
+# Toggle: enable_thinking = true|false. Thinking content streams on
+# `reasoning_content`. Chat Completions `reasoning_effort` accepts the string
+# ladder documented for the DeepSeek-V4 series (low, high, max); the gateway's
+# own validation message additionally names minimal, medium, xhigh and ultra
+# as accepted strings. Integer effort values (1..100) appear in the API
+# reference for this model but are rejected by the Token Plan endpoint.
+# Structured output: response_format json_object works (probed 2026-09-15
+# on the plan gateway); strict json_schema is rejected with 400
+# "This response_format type is unavailable now" on this host.
+# Model availability: Token Plan model list (international, Singapore region,
+# probed via the authenticated /models endpoint on Team and Personal keys).
+# Sources:
+# https://docs.qwencloud.com/api-reference/chat/openai-chat
+# https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview
+base_model = "deepseek/deepseek-v4.1-flash"
+
+reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "high", "max"] }]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0
+cache_write = 0

From 15f869b733882f6a552bd362beb524adeab01497 Mon Sep 17 00:00:00 2001
From: Oliver Mee <102673257+oliver-mee@users.noreply.github.com>
Date: Tue, 15 Sep 2026 15:49:12 +0800
Subject: [PATCH 2/2] fix(alibaba-token-plan): explicit structured_output=false
 (omission inherits the lab true)

---
 providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml b/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml
index 57cf02bed6c..19a121e6828 100644
--- a/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml
+++ b/providers/alibaba-token-plan/models/deepseek-v4.1-flash.toml
@@ -7,13 +7,15 @@
 # reference for this model but are rejected by the Token Plan endpoint.
 # Structured output: response_format json_object works (probed 2026-09-15
 # on the plan gateway); strict json_schema is rejected with 400
-# "This response_format type is unavailable now" on this host.
+# "This response_format type is unavailable now" on this host. structured_output
+# is explicitly false below: omitting it would INHERIT the lab's true value.
 # Model availability: Token Plan model list (international, Singapore region,
 # probed via the authenticated /models endpoint on Team and Personal keys).
 # Sources:
 # https://docs.qwencloud.com/api-reference/chat/openai-chat
 # https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview
 base_model = "deepseek/deepseek-v4.1-flash"
+structured_output = false
 
 reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "high", "max"] }]