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
MiniMax-M3 no longer shows model thinking/reasoning output and behavior changed #31569
Hi, I’m using MiniMax-M3 with opencode. Previously, I was able to see the model’s thinking/reasoning-style output while it was working on a task, including how it was approaching the problem.
Recently, this stopped working. Now I only see the final/normal response, and the model behavior also feels different compared to before.
I’m not sure whether this is caused by:
an opencode update
a MiniMax API/provider change
a config issue
reasoning/thinking output being disabled or no longer supported
the provider adapter not passing through the thinking/reasoning content
Expected behavior
When using MiniMax-M3 in opencode, I expected to see the same thinking/reasoning/progress output that was visible before, or at least some indication of how the model is approaching the task.
Actual behavior
The thinking/reasoning output is no longer shown. The response behavior also seems different from previous usage.
Steps to reproduce
Configure MiniMax-M3 as a provider in opencode.
Run a task that previously showed model thinking/progress.
Observe that the thinking/reasoning output is no longer visible.
Note: I’m using the API key through environment variables, not hardcoding it in the config.
Environment
OS: macOS Tahoe / Apple Silicon
opencode version:1 .16.2
Node/Bun version:v22.18.0
MiniMax model: MiniMax-M3
Provider package: @ai-sdk/anthropic
Question
Is this expected behavior now, or is there a config/provider setting required to enable the thinking/reasoning output again?
If this is caused by a provider/API change, it would be helpful to document whether MiniMax-M3 still supports visible reasoning/thinking output inside opencode.
If your issue is distinct from these (e.g., it was previously working and recently regressed), please add that context so maintainers can differentiate it.
This was fixed by #31426 and shipped in OpenCode v1.17.0. That PR enabled thinking: { "type": "adaptive" } for MiniMax M3 on the direct Anthropic-compatible route and added the none/thinking variants. This report was filed against v1.16.2, before the fix was released.
The custom provider configuration in the report also uses https://api.minimax.io as its Anthropic baseURL; MiniMax documents https://api.minimax.io/anthropic, and OpenCode's built-in provider already uses the correct Anthropic endpoint. The built-in provider should be used without overriding that URL.
For completeness, #38330 later corrected MiniMax M3 variant bodies specifically for NVIDIA and Lilac. It intentionally preserves the direct MiniMax behavior added by #31426.
Activity
github-actions commentedon Jun 9, 2026
This issue might be a duplicate of existing issues. Please check:
If your issue is distinct from these (e.g., it was previously working and recently regressed), please add that context so maintainers can differentiate it.
rekram1-node commentedon Jun 23, 2026
Automated: closing question issues.
GitHub Issues are for bug reports and feature requests only — questions / support are no longer tracked here.
If this is a bug or feature request, please open a new issue with the appropriate template.
rekram1-node commentedon Jun 23, 2026
Reopening – this was incorrectly closed by automated question-issue cleanup. Sorry for the noise.
rekram1-node commentedon Jul 22, 2026
This was fixed by #31426 and shipped in OpenCode v1.17.0. That PR enabled
thinking: { "type": "adaptive" }for MiniMax M3 on the direct Anthropic-compatible route and added thenone/thinkingvariants. This report was filed against v1.16.2, before the fix was released.The custom provider configuration in the report also uses
https://api.minimax.ioas its AnthropicbaseURL; MiniMax documentshttps://api.minimax.io/anthropic, and OpenCode's built-in provider already uses the correct Anthropic endpoint. The built-in provider should be used without overriding that URL.For completeness, #38330 later corrected MiniMax M3 variant bodies specifically for NVIDIA and Lilac. It intentionally preserves the direct MiniMax behavior added by #31426.
Closing as completed.