Repository navigation
[Question]: OpenAI reasoning effort fails with Unknown parameter: 'thinking' #94
Description
Preflight checklist
- I checked the README (Configuration + Troubleshooting) and searched existing issues.
Your question
OpenAI reasoning effort fails with Unknown parameter: 'thinking'.
When using the Prompt Optimizer with the OpenAI provider, any non-off reasoning effort causes the request to fail with:
API 400:
{
"error": {
"message": "Unknown parameter: 'thinking'.",
"type": "invalid_request_error",
"param": "thinking",
"code": "unknown_parameter"
}
}
If I set reasoning effort to off, the optimizer works normally.
Is the OpenAI adapter incorrectly sending a thinking parameter when reasoning is enabled? What is the correct way to enable reasoning effort for OpenAI models?
What have you tried?
I tested the following configuration:
- API format: openai
- API URL: https://api.openai.com/v1
- Model: gpt-5.6-sol
- Reasoning effort: max
With max enabled, I consistently receive the Unknown parameter: 'thinking' 400 error.
I also tried setting reasoning effort to off. With off, the same API key, model, URL, and prompt work successfully.
This makes it look like the issue is specifically related to the parameter sent when reasoning is enabled rather than API authentication or general connectivity.
Context
Windows 11 x64, using the OpenAI API directly (no proxy).
- Extension: Claude Code Usage
- Extension ID: GrowthJack.claude-code-usage
- Extension version: 2.3.0
- Publisher: GrowthJack
- VS Code version: 1.135.0 (user setup)
- VS Code commit: 08d4889f9ec4a1685d257b9b95de036c8e1ce1e5
- OS: Windows_NT x64 10.0.26200
- Provider: OpenAI
- Endpoint: https://api.openai.com/v1
- Model: gpt-5.6-sol
- Reasoning effort
max: fails withUnknown parameter: 'thinking' - Reasoning effort
off: works normally - No proxy / CC Switch / third-party API gateway
Activity
github-actions commentedon Aug 28, 2026
🤖 Automated first-pass reply
The configured analysis model did not return a usable analysis for this issue. The item has still been received and remains available for maintainer review.
To help with triage, please ensure the extension, VS Code, and operating-system versions plus anonymous diagnostic logs are included.
🤖 Posted automatically by CCU Bot — not a maintainer decision.
Alex668866 commentedon Aug 31, 2026
You’re right — this is a provider payload bug, not an authentication or model-support issue.
At this point in the OpenAI chat-completions branch, the request adds both
reasoning_effortandthinking. OpenAI Chat Completions accepts the top-levelreasoning_effort;thinkingis not a valid OpenAI parameter, which explains the 400. GPT-5.6 Sol does supportmax.Removing
body.thinkingwhenapiFormat === "openai"should fix it. I’d also add a request-body test asserting that OpenAI payloads never containthinking. If this moves to the Responses API later, the equivalent shape isreasoning: { effort: "max" }.Carl723000 commentedon Sep 2, 2026
Confirmed; thanks @Alex668866 for the precise diagnosis. The OpenAI-compatible request builder now emits reasoning_effort without the unsupported top-level thinking field. Regression tests cover both the DeepSeek-style path and a direct OpenAI GPT-5.6 Sol path, including max effort. The v2.3.1 release candidate passes the full local unit and UI suites plus an installed-VSIX smoke test. I am keeping this issue open until the fix is pushed and released, then it can be closed against the published build.
Carl723000 commentedon Sep 6, 2026
Fixed on
mainin087bae5and included in the reviewed v2.3.1 draft.The OpenAI-compatible Chat Completions request now sends
reasoning_effortwithout the unsupported top-levelthinkingfield. Regression coverage includes the direct OpenAI GPT-5.6 Sol path withmaxeffort and the existing DeepSeek-style path. The release PR passed Node, UI, and packaged-VSIX checks.Closing as implemented; the maintainer-controlled v2.3.1 Release remains a draft until Publish is clicked.
🤖 Generated with OpenAI Codex