gpt 4.1 max_tokens too large

[dvic]

│Message:                                                                                                                                  │
│  POST "https://api.openai.com/v1/chat/completions": 400 Bad Request {                                                                    │
│      "message": "max_tokens is too large: 50000. This model supports at most 32768 completion tokens, whereas you provided 50000.",      │
│      "type": "invalid_request_error",                                                                                                    │
│      "param": "max_tokens",                                                                                                              │
│      "code": "invalid_value"                                                                                                             │
│    }                                                                                                                                     │
│                                                                                                                                          │
│

[rekram1-node]

can you provide your config and opencode version? @dvic

[dvic]

No custom config, just OPENAI_API_KEY, will check version and report

[rekram1-node]

I was able to replicate this when I accessed a previous session and tried asking follow ups, will investigate further

[dvic]

I was able to replicate this when I accessed a previous session and tried asking follow ups, will investigate further

Hmm, I get it immediately when I'm asked to init the project.

[rekram1-node]

good to know, I will investigate that as well, ty @dvic

[jdlms]

Also ran into this. Claude and his 50000 max_tokens is being always set as the default model when api key is set as env var with no config files. Opened pr.

[rekram1-node]

Thanks for you help! I couldnt find the time to look into it this weekend

[dvic]

Can confirm that #147 fixes the issue!

[WarcraftYax]

@dvic Do you know if this fix made it into a release? I can see it was merged but can't see it in the release notes of any release and I'm experiencing this issue.

[dvic]

It should be in this release: https://github.com/opencode-ai/opencode/releases/tag/v0.0.49