You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
GPT-5.5 with tools on custom OpenAI-compatible providers still lacks a clear working path (chat/completions vs responses) #35732
I'm using OpenCode with a custom OpenAI-compatible provider in an enterprise setup.
I now have a confirmed working path for o4-mini reasoning:
plain model name (o4-mini)
no reasoning suffix in the model name
agent-level options.reasoning_effort
tool calling works
However, gpt-5.5 with tools is still unclear/problematic on custom OpenAI-compatible providers.
The backend error is:
Function tools with reasoning_effort are not supported for gpt-5.5 in /v1/chat/completions. Please use /v1/responses instead.
Upstream issues such as #26219 and #29325 were closed, but from a user perspective there is still no clear/documented happy path for this setup when tools are enabled.
What I would expect
One of the following should be true and documented:
OpenCode should automatically avoid sending incompatible reasoning_effort combinations for GPT-5.5 when using chat/completions with tools, OR
OpenCode should correctly support/route the responses path for compatible custom providers, OR
OpenCode should clearly document that GPT-5.5 + tools is unsupported on custom OpenAI-compatible providers unless a specific provider adapter/path is used.
Why this matters
Today the behavior is hard to reason about when integrating non-default gateways:
plain model usage may work
tool usage may fail
the failure mode depends on endpoint semantics
the supported configuration is not obvious from the config surface
This creates ambiguity for enterprise/custom-provider users.
Environment
OpenCode version: 1.17.4
Provider type: custom OpenAI-compatible provider
Model: GPT-5.5
Tools enabled: yes
Additional context
In the same environment, o4-mini works correctly when configured as:
plain model name
no suffix notation
agent-level options.reasoning_effort
tools enabled
So this seems specifically about GPT-5.5 + tools + endpoint compatibility, not a general custom-provider failure.
If your issue is distinct from all of these (e.g. the prior closed issues did not fully resolve the GPT-5.5 + tools + custom provider path), please add a note clarifying what remains unresolved after those fixes.
Activity
github-actions commentedon Jul 7, 2026
This issue might be a duplicate of existing issues. Please check:
options(reasoning/thinking) not forwarded to API for@ai-sdk/openai-compatibleproviders in headless mode #27361: Modeloptions(reasoning/thinking) not forwarded to API for@ai-sdk/openai-compatibleproviders in headless mode (open, related reasoning_effort + custom provider issue)If your issue is distinct from all of these (e.g. the prior closed issues did not fully resolve the GPT-5.5 + tools + custom provider path), please add a note clarifying what remains unresolved after those fixes.
github-actions commentedon Sep 6, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.