`meat: model generate: openai stream error :`

[zg]

Hi - congrats, this appears to be your first issue! 😀

I just installed meat using the instructions on the README, but ran into a strange issue:

zzg@zzghl secret-project % meat
meat: model generate: openai stream error :

I am unsure how to proceed - the base URL is https://api.openai.com/v1.

[joshi4]

Ran into the same issue.

Using Anthropic models worked for me.

meat -model claude-opus-4-8 (requires ant api key in your env)

[obakeng-develops]

Same error here, on meat.dev v0.0.0-20260803201634-f39f41dfe7b5, go1.26.4, darwin/arm64.

The key and the account check out. GET /v1/models returns 200 across 118 models. Streaming works when I call the API directly: POST /v1/responses with gpt-5, stream: true and reasoning: {"effort": "low"} opens a stream and returns 200. Adding "summary": "auto" also works, so this is not an org verification limit on reasoning summaries.

Through meat, every reasoning model fails with the same empty message: gpt-5, gpt-5-mini, gpt-5.1.

Non-reasoning models fail differently, with a clean 400:

meat: model generate: openai API 400: {
  "error": {
    "message": "Unsupported parameter: 'reasoning.effort' is not supported with this model.",
    "code": "unsupported_parameter"

So reasoning.effort is sent unconditionally. That rules out gpt-4o and gpt-4.1. On models that do accept it, the stream fails for a request shape the API accepts from curl.

Diff size is not involved. A one line commit fails the same as a 700 line range.

The empty string after stream error : suggests the underlying error is being swallowed. Surfacing it would probably identify the cause.