Repository navigation
Non-DeepSeek custom providers fail when reasoning_content is sent in tool calls or subsequent messages #37475
Closed as not planned
Description
Description
The provider fails because not all OpenAI-compatible endpoints recognize or accept the reasoning_content parameter.
Plugins
No Plugins used
OpenCode version
1.18.3
Steps to reproduce
- Add an OpenAI-compatible custom provider (e.g., Cerebras GPT-OSS-120B).
- Start a new session using this non-DeepSeek model and send a message.
- Scenario A: If the message does not trigger a tool call or MCP server, the response and thinking contents display correctly.
- Scenario B: If a tool call is initiated, the request includes reasoning_content, causing the provider to fail.
- Scenario C: If the first message succeeds without error, send a second message. This will 100% fail because the provider receives reasoning_content in the request history.
Screenshot and/or share link
Operating System
Windows 11
Terminal
Windows Terminal
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Jul 17, 2026
This issue might be a duplicate of existing issues. Please check:
These all relate to the same underlying problem: reasoning_content being incorrectly included or excluded from provider history depending on whether the provider supports it.
github-actions commentedon Sep 17, 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.