What happened?
Using a GPT-5.6 model (e.g. gpt-5.6-terra) on the OpenAI endpoint with an Agent that has tools (MCP servers, or any function tools) and a reasoning effort set fails on every request with:
Something went wrong. Here's the specific error message we encountered:
An error occurred while processing the request: 400 Function tools with reasoning_effort
are not supported for gpt-5.6-terra in /v1/chat/completions. To use function tools,
use /v1/responses or set reasoning_effort to 'none'.
Expected: the request succeeds — reasoning and function tools should work together on GPT-5.6 models, as they do on the Responses API.
Why it happens: OpenAI changed behavior for GPT-5.6 models: /v1/chat/completions now rejects requests that combine function tools with reasoning_effort other than 'none' (OpenAI community report). LibreChat defaults to Chat Completions (useResponsesApi defaults to false) and sends flat reasoning_effort there (packages/api/src/endpoints/openai/llm.ts, applyReasoningConfig → llmConfig.reasoning_effort). The reasoning-model regex /\b(o[13]|gpt-5)(?!\.|-chat)(?:-|$)/ intentionally excludes dotted versions, so gpt-5.6-* gets no special handling, and nothing switches these models to the Responses API automatically (only web_search does that today).
Users can work around it per conversation/agent by manually enabling the "Use Responses API" toggle, but out of the box every GPT-5.6 + tools + reasoning request fails.
Version Information
Reproduced on a v0.8.6-based deployment. Confirmed by code inspection and unit tests that the code path is unchanged on current main (cf9a426).
Steps to Reproduce
- Configure the OpenAI endpoint with a GPT-5.6 model (e.g.
gpt-5.6-terra)
- Create an Agent using that model, attach any tool (e.g. an MCP server), and set Reasoning Effort to any value other than
none (leaving the "Use Responses API" toggle untouched/off)
- Send any message
- The request fails with the 400 error above
What browsers are you seeing the problem on?
No response (server-side issue, browser-independent)
Relevant log output
400 Function tools with reasoning_effort are not supported for gpt-5.6-terra in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to 'none'.
Screenshots
No response
What happened?
Using a GPT-5.6 model (e.g.
gpt-5.6-terra) on the OpenAI endpoint with an Agent that has tools (MCP servers, or any function tools) and a reasoning effort set fails on every request with:Expected: the request succeeds — reasoning and function tools should work together on GPT-5.6 models, as they do on the Responses API.
Why it happens: OpenAI changed behavior for GPT-5.6 models:
/v1/chat/completionsnow rejects requests that combine functiontoolswithreasoning_effortother than'none'(OpenAI community report). LibreChat defaults to Chat Completions (useResponsesApidefaults tofalse) and sends flatreasoning_effortthere (packages/api/src/endpoints/openai/llm.ts,applyReasoningConfig→llmConfig.reasoning_effort). The reasoning-model regex/\b(o[13]|gpt-5)(?!\.|-chat)(?:-|$)/intentionally excludes dotted versions, sogpt-5.6-*gets no special handling, and nothing switches these models to the Responses API automatically (onlyweb_searchdoes that today).Users can work around it per conversation/agent by manually enabling the "Use Responses API" toggle, but out of the box every GPT-5.6 + tools + reasoning request fails.
Version Information
Reproduced on a v0.8.6-based deployment. Confirmed by code inspection and unit tests that the code path is unchanged on current
main(cf9a426).Steps to Reproduce
gpt-5.6-terra)none(leaving the "Use Responses API" toggle untouched/off)What browsers are you seeing the problem on?
No response (server-side issue, browser-independent)
Relevant log output
Screenshots
No response