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
[FEATURE]: support thinking mode variants for MiniMax M3 #31180
MiniMax M3 supports two different thinking modes. The default is thinking mode, we can add non-thinking mode as a variant.
FYI I tried to add an variant as follows. The variant itself was added correctly, I can switch to it, but thinking was not disabled. Seems like the way I specified the option was wrong.
Without "reasoning":true,the option thinking does not seem to work.
I strongly agree that this support should be added.
By the way, while filling in the settings, I noticed that the OpenCode schema seems to be broken—the model.options field lacks the corresponding schema hints. Has anyone else encountered this issue?
Thank you for sharing the setting. Unfortunately I tried it but it still generated thinking tokens. Tested on OpenCode 1.16.2
Well, I did some more tests, and it seems that this may be because you didn't fill in npm, baseUrl, and apiKey fields, which led to this not being a valid provider setup. At the same time, you reused the built-in name minimax-coding-plan, so OpenCode automatically fell back to its built-in MiniMax-M3 model settings instead of the one you set. I'm not entirely sure if this reasoning is correct, but it is consistent with the results of my tests.
Activity
qiyuanhuakai commentedon Jun 7, 2026
I have tested this and found that the corret option is
Without
"reasoning":true,the optionthinkingdoes not seem to work.I strongly agree that this support should be added.
By the way, while filling in the settings, I noticed that the OpenCode schema seems to be broken—the
model.optionsfield lacks the corresponding schema hints. Has anyone else encountered this issue?lucidfrontier45 commentedon Jun 7, 2026
@qiyuanhuakai
Thank you for sharing the setting. Unfortunately I tried it but it still generated thinking tokens.
Tested on OpenCode 1.16.2
qiyuanhuakai commentedon Jun 7, 2026
Well, I did some more tests, and it seems that this may be because you didn't fill in
npm,baseUrl, andapiKeyfields, which led to this not being a valid provider setup. At the same time, you reused the built-in nameminimax-coding-plan, so OpenCode automatically fell back to its built-inMiniMax-M3model settings instead of the one you set. I'm not entirely sure if this reasoning is correct, but it is consistent with the results of my tests.lucidfrontier45 commentedon Jun 7, 2026
@qiyuanhuakai
Yes. That's my intention. I wanted to override built-in minimax-coding-plan.
I followed the official doc to do that.
https://opencode.ai/docs/en/models/#configure-models