anomalyco · rekram1-node · Jun 26, 2026 · Jun 26, 2026
diff --git a/providers/302ai/provider.toml b/providers/302ai/provider.toml
@@ -1,5 +1,11 @@
 name = "302.AI"
 env = ["302AI_API_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Audited POST https://api.302.ai/v1/chat/completions. The provider's API guide
+# documents model/messages only; no reasoning toggle, effort, or numeric budget
+# request field is documented. Do not infer passthrough from upstream APIs.
+# Sources:
+# https://doc.302.ai/
 doc = "https://doc.302.ai"
 api = "https://api.302.ai/v1"
diff --git a/providers/abacus/provider.toml b/providers/abacus/provider.toml
@@ -1,5 +1,11 @@
 name = "Abacus"
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Audited POST https://routellm.abacus.ai/v1/chat/completions. The provider API
+# reference documents no reasoning toggle, effort, or numeric budget request
+# field. Do not infer behavior from the routed model developer's API.
+# Sources:
+# https://abacus.ai/help/api
 env = ["ABACUS_API_KEY"]
 doc = "https://abacus.ai/help/api"
 api = "https://routellm.abacus.ai/v1"
diff --git a/providers/abliteration-ai/models/abliterated-model.toml b/providers/abliteration-ai/models/abliterated-model.toml
@@ -1,4 +1,10 @@
 name = "Abliterated Model"
+# Reasoning HTTP format (accessed 2026-06-25):
+# This model thinks by default. On POST /v1/chat/completions or /v1/messages,
+# top-level `thinking: false` skips thinking; omission keeps it enabled.
+# Sources:
+# https://docs.abliteration.ai/models
+# https://docs.abliteration.ai/capabilities/thinking
 release_date = "2026-01-06"
 last_updated = "2026-01-06"
 attachment = true

diff --git a/providers/abliteration-ai/provider.toml b/providers/abliteration-ai/provider.toml
@@ -1,5 +1,12 @@
 name = "abliteration.ai"
 env = ["ABLIT_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# POST /v1/chat/completions and POST /v1/messages: top-level `thinking` is true
+# by default; false skips thinking. POST /v1/responses has no thinking toggle.
+# No effort or numeric reasoning-budget request field is documented.
+# Sources:
+# https://docs.abliteration.ai/capabilities/thinking
+# https://docs.abliteration.ai/compatibility-matrix
 api = "https://api.abliteration.ai/v1"
 doc = "https://docs.abliteration.ai/models"
diff --git a/providers/aihubmix/models/claude-opus-4-6.toml b/providers/aihubmix/models/claude-opus-4-6.toml
@@ -5,6 +5,7 @@ last_updated = "2026-03-13"
 attachment = true
 reasoning = true
 reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }]
+# Native Messages prefers $.thinking.type = "adaptive" with $.output_config.effort = "low"|"medium"|"high"|"max"; enabled budget_tokens >= 1024 is deprecated and must be < $.max_tokens. https://docs.aihubmix.com/cn/api/Claude-Native (accessed 2026-06-25)
 temperature = true
 tool_call = true
 structured_output = true

diff --git a/providers/aihubmix/models/claude-opus-4-7.toml b/providers/aihubmix/models/claude-opus-4-7.toml
@@ -5,6 +5,7 @@ last_updated = "2026-04-16"
 attachment = true
 reasoning = true
 reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high", "xhigh", "max"] }]
+# Native Messages uses $.thinking.type = "adaptive" and $.output_config.effort = "low"|"medium"|"high"|"xhigh"|"max"; manual budget_tokens is rejected. https://docs.aihubmix.com/cn/blogs/Claude-Opus4.7 (accessed 2026-06-25)
 temperature = false
 tool_call = true
 structured_output = true

diff --git a/providers/aihubmix/models/claude-sonnet-4-6.toml b/providers/aihubmix/models/claude-sonnet-4-6.toml
@@ -5,6 +5,7 @@ last_updated = "2026-03-13"
 attachment = true
 reasoning = true
 reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high", "max"] }, { type = "budget_tokens", min = 1_024 }]
+# Native Messages prefers $.thinking.type = "adaptive" with $.output_config.effort = "low"|"medium"|"high"; enabled budget_tokens >= 1024 is deprecated and must be < $.max_tokens. Chat effort "max" maps to native "high". https://docs.aihubmix.com/cn/api/Claude-Native (accessed 2026-06-25)
 temperature = true
 tool_call = true
 structured_output = true

diff --git a/providers/aihubmix/models/gemini-2.5-flash.toml b/providers/aihubmix/models/gemini-2.5-flash.toml
@@ -5,6 +5,7 @@ last_updated = "2025-06-05"
 attachment = true
 reasoning = true
 reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }, { type = "budget_tokens", min = 0, max = 24_576 }]
+# Native Gemini uses $.generationConfig.thinkingConfig.thinkingBudget: 0 disables, -1 is dynamic, and manual budgets are 1..24576. https://cloud.google.com/vertex-ai/generative-ai/docs/thinking (accessed 2026-06-25)
 temperature = true
 tool_call = true
 structured_output = true

diff --git a/providers/aihubmix/models/gemini-2.5-pro.toml b/providers/aihubmix/models/gemini-2.5-pro.toml
@@ -5,6 +5,7 @@ last_updated = "2025-06-05"
 attachment = true
 reasoning = true
 reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+# Native Gemini uses $.generationConfig.thinkingConfig.thinkingBudget: -1 is dynamic and manual budgets are 128..32768; 0/off is unsupported. https://cloud.google.com/vertex-ai/generative-ai/docs/thinking (accessed 2026-06-25)
 temperature = true
 tool_call = true
 structured_output = true

diff --git a/providers/aihubmix/provider.toml b/providers/aihubmix/provider.toml
@@ -1,5 +1,9 @@
 name = "AIHubMix"
 npm = "@aihubmix/ai-sdk-provider"
+# Raw Chat: $.reasoning_effort = "none"|"minimal"|"low"|"medium"|"high"|"xhigh"; aliases are $.reasoning.effort and integer $.reasoning.max_tokens. "none" disables models that permit it. https://docs.aihubmix.com/cn/api/unified-inference (accessed 2026-06-25)
+# Raw Responses: $.reasoning.effort carries effort; this endpoint has no reasoning-token budget field. https://docs.aihubmix.com/cn/api-reference/openai-compatible/create-a-model-response (accessed 2026-06-25)
+# Raw Messages: $.thinking.type = "enabled"|"disabled"|"adaptive"; enabled uses $.thinking.budget_tokens >= 1024, and $.output_config.effort = "low"|"medium"|"high"|"xhigh"|"max" subject to model support. https://docs.aihubmix.com/cn/api-reference/anthropic-compatible/create-a-message (accessed 2026-06-25)
+# Raw Gemini native: $.generationConfig.thinkingConfig uses integer thinkingBudget (-1 dynamic; 0 off where supported) or string thinkingLevel; model bounds differ below. https://docs.aihubmix.com/cn/api-reference/google-vertex-ai-compatible/generate-content (accessed 2026-06-25)
 env = ["AIHUBMIX_API_KEY"]
 doc = "https://docs.aihubmix.com"
 
diff --git a/providers/alibaba-cn/provider.toml b/providers/alibaba-cn/provider.toml
@@ -1,5 +1,20 @@
 name = "Alibaba (China)"
 env = ["DASHSCOPE_API_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# OpenAI Chat: POST /compatible-mode/v1/chat/completions; top-level
+# `enable_thinking` is true or false and `thinking_budget` is an integer token
+# cap (Qwen3 in thinking mode and Kimi only; default is the model maximum).
+# Responses: POST /compatible-mode/v1/responses; `reasoning.effort` is none,
+# minimal, low, medium (default), or high; no numeric thinking budget is accepted.
+# Anthropic: POST /apps/anthropic/v1/messages; `thinking.type` is enabled or
+# disabled and `thinking.budget_tokens` is an integer used only when enabled.
+# `reasoning_effort` is high or max only for DeepSeek V4 Pro/Flash.
+# DashScope text: POST /api/v1/services/aigc/text-generation/generation; put
+# `enable_thinking` and `thinking_budget` under `parameters`, not at JSON root.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 doc = "https://www.alibabacloud.com/help/en/model-studio/models"
 api = "https://dashscope.aliyuncs.com/compatible-mode/v1" 
diff --git a/providers/alibaba-coding-plan-cn/models/MiniMax-M2.5.toml b/providers/alibaba-coding-plan-cn/models/MiniMax-M2.5.toml
@@ -1,4 +1,10 @@
 name = "MiniMax-M2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model; Alibaba-deployed MiniMax-M2.5 is
+# thinking-only. No toggle, effort, or numeric budget request field is documented.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
+# https://help.aliyun.com/zh/model-studio/deep-thinking
 family = "minimax"
 release_date = "2026-02-12"
 last_updated = "2026-02-12"

diff --git a/providers/alibaba-coding-plan-cn/models/glm-5.toml b/providers/alibaba-coding-plan-cn/models/glm-5.toml
@@ -1,4 +1,10 @@
 name = "GLM-5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking is on by default and
+# top-level `enable_thinking`: true|false toggles it. No effort/budget documented.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
+# https://help.aliyun.com/zh/model-studio/deep-thinking
 family = "glm"
 release_date = "2026-02-11"
 last_updated = "2026-02-11"

diff --git a/providers/alibaba-coding-plan-cn/models/kimi-k2.5.toml b/providers/alibaba-coding-plan-cn/models/kimi-k2.5.toml
@@ -1,4 +1,11 @@
 name = "Kimi K2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking uses top-level
+# `enable_thinking`: true enables; false (default) disables. No effort or numeric
+# budget bound is documented specifically for the plan endpoint.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/kimi-api
 family = "kimi-k2"
 release_date = "2026-01-27"
 last_updated = "2026-01-27"

diff --git a/providers/alibaba-coding-plan-cn/models/qwen3-coder-plus.toml b/providers/alibaba-coding-plan-cn/models/qwen3-coder-plus.toml
@@ -1,4 +1,9 @@
 name = "Qwen3 Coder Plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. The plan docs document no reasoning
+# toggle, effort, or numeric budget field for this model.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
 family = "qwen"
 release_date = "2025-07-23"
 last_updated = "2025-07-23"

diff --git a/providers/alibaba-coding-plan-cn/models/qwen3.7-plus.toml b/providers/alibaba-coding-plan-cn/models/qwen3.7-plus.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.7-plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking is on by default;
+# top-level `enable_thinking`: true|false toggles it. The plan docs do not state
+# a numeric budget bound or a plan-specific effort control.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
+# https://help.aliyun.com/zh/model-studio/deep-thinking
 reasoning_options = [{ type = "toggle" }]
 
 [cost]

diff --git a/providers/alibaba-coding-plan-cn/provider.toml b/providers/alibaba-coding-plan-cn/provider.toml
@@ -1,5 +1,15 @@
 name = "Alibaba Coding Plan (China)"
 env = ["ALIBABA_CODING_PLAN_API_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# OpenAI plan base https://coding.dashscope.aliyuncs.com/v1 uses POST
+# /chat/completions with top-level `enable_thinking`: true or false.
+# Anthropic plan base https://coding.dashscope.aliyuncs.com/apps/anthropic uses
+# POST /v1/messages with `thinking.type`: enabled or disabled and optional
+# integer `thinking.budget_tokens`. No plan-specific effort field is documented.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/coding-plan
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 doc = "https://help.aliyun.com/zh/model-studio/coding-plan"
 api = "https://coding.dashscope.aliyuncs.com/v1"
diff --git a/providers/alibaba-coding-plan/models/MiniMax-M2.5.toml b/providers/alibaba-coding-plan/models/MiniMax-M2.5.toml
@@ -1,4 +1,10 @@
 name = "MiniMax-M2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model; Alibaba-deployed MiniMax-M2.5 is
+# thinking-only. No toggle, effort, or numeric budget request field is documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 family = "minimax"
 release_date = "2026-02-12"
 last_updated = "2026-02-12"

diff --git a/providers/alibaba-coding-plan/models/glm-5.toml b/providers/alibaba-coding-plan/models/glm-5.toml
@@ -1,4 +1,10 @@
 name = "GLM-5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking is on by default and
+# top-level `enable_thinking`: true|false toggles it. No effort/budget documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 family = "glm"
 release_date = "2026-02-11"
 last_updated = "2026-02-11"

diff --git a/providers/alibaba-coding-plan/models/kimi-k2.5.toml b/providers/alibaba-coding-plan/models/kimi-k2.5.toml
@@ -1,4 +1,11 @@
 name = "Kimi K2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking uses top-level
+# `enable_thinking`: true enables; false (default) disables. No effort or numeric
+# budget bound is documented specifically for the plan endpoint.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/kimi-api
 family = "kimi-k2"
 release_date = "2026-01-27"
 last_updated = "2026-01-27"

diff --git a/providers/alibaba-coding-plan/models/qwen3-coder-plus.toml b/providers/alibaba-coding-plan/models/qwen3-coder-plus.toml
@@ -1,4 +1,9 @@
 name = "Qwen3 Coder Plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. The plan docs document no reasoning
+# toggle, effort, or numeric budget field for this model.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
 family = "qwen"
 release_date = "2025-07-23"
 last_updated = "2025-07-23"

diff --git a/providers/alibaba-coding-plan/models/qwen3.7-plus.toml b/providers/alibaba-coding-plan/models/qwen3.7-plus.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.7-plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Coding Plan exact-string allowlist model. Hybrid thinking is on by default;
+# top-level `enable_thinking`: true|false toggles it. The plan docs do not state
+# a numeric budget bound or a plan-specific effort control.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 reasoning_options = [{ type = "toggle" }]
 
 [cost]

diff --git a/providers/alibaba-coding-plan/provider.toml b/providers/alibaba-coding-plan/provider.toml
@@ -1,5 +1,15 @@
 name = "Alibaba Coding Plan"
 env = ["ALIBABA_CODING_PLAN_API_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# OpenAI plan base https://coding-intl.dashscope.aliyuncs.com/v1 uses POST
+# /chat/completions with top-level `enable_thinking`: true or false.
+# Anthropic plan base https://coding-intl.dashscope.aliyuncs.com/apps/anthropic
+# uses POST /v1/messages with `thinking.type`: enabled or disabled and optional
+# integer `thinking.budget_tokens`. No plan-specific effort field is documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/coding-plan
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 doc = "https://www.alibabacloud.com/help/en/model-studio/coding-plan"
 api = "https://coding-intl.dashscope.aliyuncs.com/v1"
diff --git a/providers/alibaba-token-plan-cn/models/MiniMax-M2.5.toml b/providers/alibaba-token-plan-cn/models/MiniMax-M2.5.toml
@@ -1,4 +1,9 @@
 base_model = "minimax/MiniMax-M2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Alibaba-deployed MiniMax-M2.5 is thinking-only. No toggle, effort, or numeric
+# reasoning-budget request field is documented for this model.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
 reasoning_options = []
 
 [interleaved]

diff --git a/providers/alibaba-token-plan-cn/models/deepseek-v4-flash.toml b/providers/alibaba-token-plan-cn/models/deepseek-v4-flash.toml
@@ -1,4 +1,11 @@
 base_model = "deepseek/deepseek-v4-flash"
+# Reasoning HTTP format (accessed 2026-06-25):
+# This model toggles with `enable_thinking`: true or false. Chat Completions
+# `reasoning_effort` accepts high (default) or max; low/medium map to high and
+# xhigh maps to max. Anthropic Messages defaults `reasoning_effort` to max.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/deepseek-api
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/models/deepseek-v4-pro.toml b/providers/alibaba-token-plan-cn/models/deepseek-v4-pro.toml
@@ -1,4 +1,11 @@
 base_model = "deepseek/deepseek-v4-pro"
+# Reasoning HTTP format (accessed 2026-06-25):
+# This model toggles with `enable_thinking`: true or false. Chat Completions
+# `reasoning_effort` accepts high (default) or max; low/medium map to high and
+# xhigh maps to max. Anthropic Messages defaults `reasoning_effort` to max.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/deepseek-api
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/models/kimi-k2.5.toml b/providers/alibaba-token-plan-cn/models/kimi-k2.5.toml
@@ -1,4 +1,10 @@
 base_model = "moonshotai/kimi-k2.5"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid model: top-level `enable_thinking` true enables thinking; false is the
+# default. `thinking_budget` is an integer token cap; no bounds are documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/kimi-api
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 base_model_omit = ["structured_output"]
 family = "kimi-k2"
 attachment = true

diff --git a/providers/alibaba-token-plan-cn/models/kimi-k2.6.toml b/providers/alibaba-token-plan-cn/models/kimi-k2.6.toml
@@ -1,4 +1,10 @@
 base_model = "moonshotai/kimi-k2.6"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid model: top-level `enable_thinking` true enables thinking; false is the
+# default. `thinking_budget` is an integer token cap; no bounds are documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/kimi-api
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 base_model_omit = ["structured_output"]
 family = "kimi-k2"
 

diff --git a/providers/alibaba-token-plan-cn/models/kimi-k2.7-code.toml b/providers/alibaba-token-plan-cn/models/kimi-k2.7-code.toml
@@ -1,4 +1,10 @@
 base_model = "moonshotai/kimi-k2.7-code"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Thinking-only model: `enable_thinking` defaults to true and cannot disable
+# thinking. `thinking_budget` is an integer token cap; no bounds are documented.
+# Sources:
+# https://www.alibabacloud.com/help/en/model-studio/kimi-api
+# https://www.alibabacloud.com/help/en/model-studio/deep-thinking
 base_model_omit = ["structured_output"]
 family = "kimi-k2"
 reasoning_options = [{ type = "toggle" }, { type = "budget_tokens" }]

diff --git a/providers/alibaba-token-plan-cn/models/qwen3.6-flash.toml b/providers/alibaba-token-plan-cn/models/qwen3.6-flash.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.6-flash"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid, thinking on by default. Chat/DashScope `enable_thinking`: true|false;
+# integer `thinking_budget` maximum 81920; no minimum or disable sentinel is
+# documented. Responses instead uses `reasoning.effort` and no numeric budget.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/models/qwen3.6-plus.toml b/providers/alibaba-token-plan-cn/models/qwen3.6-plus.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.6-plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid, thinking on by default. Chat/DashScope `enable_thinking`: true|false;
+# integer `thinking_budget` maximum 81920; no minimum or disable sentinel is
+# documented. Responses instead uses `reasoning.effort` and no numeric budget.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/models/qwen3.7-max.toml b/providers/alibaba-token-plan-cn/models/qwen3.7-max.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.7-max"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid, thinking on by default. Chat/DashScope `enable_thinking`: true|false;
+# integer `thinking_budget` maximum 262144; no minimum or disable sentinel is
+# documented. Responses instead uses `reasoning.effort` and no numeric budget.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/models/qwen3.7-plus.toml b/providers/alibaba-token-plan-cn/models/qwen3.7-plus.toml
@@ -1,4 +1,11 @@
 base_model = "alibaba/qwen3.7-plus"
+# Reasoning HTTP format (accessed 2026-06-25):
+# Hybrid, thinking on by default. Chat/DashScope `enable_thinking`: true|false;
+# integer `thinking_budget` maximum 262144; no minimum or disable sentinel is
+# documented. Responses instead uses `reasoning.effort` and no numeric budget.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
 
 [[reasoning_options]]
 type = "toggle"

diff --git a/providers/alibaba-token-plan-cn/provider.toml b/providers/alibaba-token-plan-cn/provider.toml
@@ -1,5 +1,17 @@
 name = "Alibaba Token Plan (China)"
 env = ["ALIBABA_TOKEN_PLAN_API_KEY"]
 npm = "@ai-sdk/openai-compatible"
+# Reasoning HTTP format (accessed 2026-06-25):
+# OpenAI plan base .../compatible-mode/v1 uses POST /chat/completions with
+# top-level `enable_thinking`: true or false and integer `thinking_budget`.
+# POST /responses uses `reasoning.effort`: none, minimal, low, medium, or high;
+# `thinking_budget` is not accepted there. Anthropic plan base .../apps/anthropic
+# uses POST /v1/messages with `thinking.type`: enabled or disabled, integer
+# `thinking.budget_tokens`, and DeepSeek V4-only `reasoning_effort`: high or max.
+# Sources:
+# https://help.aliyun.com/zh/model-studio/token-plan-quickstart
+# https://help.aliyun.com/zh/model-studio/deep-thinking
+# https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api
+# https://www.alibabacloud.com/help/en/model-studio/anthropic-api-messages
 doc = "https://www.alibabacloud.com/help/zh/model-studio/token-plan-overview"
 api = "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1"