[OpenCode Go] Models other than GLM-4-5 not accessible via subscription — HTTP 401 "Model not supported" error

[jacob3712]

Description

Describe the bug

I have an active OpenCode Go subscription and am accessing it via the Hermes agent on a Linux VM. Only GLM-4-5 appears to be accessible. All other models (e.g. minimax-m2.7) return an HTTP 401 error with the message Model minimax-m2-7 not supported when called against the OpenCode Go endpoint.

Endpoint
https://opencode.ai/zen/go

Expected behavior

All models advertised under the OpenCode Go subscription should be accessible and return valid responses.

Actual behavior

Requests to models other than GLM-4-5 via the Go subscription endpoint return:

HTTP 401: Model minimax-m2-7 not supported

The error is non-retryable and the agent aborts. GLM-4-5 is the only model that works.

Steps to reproduce

1. Set up OpenCode Go subscription and configure the API key
2. Point Hermes agent (or any OpenAI-compatible client) to https://opencode.ai/zen/go
3. Send a request using a model other than GLM-4-5 (e.g. minimax-m2.7)
4. Observe HTTP 401 response: Model minimax-m2-7 not supported

Environment

• OS: Linux (VM)
• Client: Hermes agent
• OpenCode Go subscription: Active
• Models affected: All except GLM-4-5 (confirmed broken: minimax-m2.7)
• Provider configured as: opencode-go

Additional context

The 401 response body is:

{"type": "error", "error": {"type": "ModelError", "message": "Model minimax-m2-7 not supported"}}

This appears to be a backend model availability/routing issue on the Go subscription tier rather than an authentication problem, since GLM-4-5 works fine with the same credentials.

Plugins

No response

OpenCode version

No response

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[lxcalghost]

It's most likely not a problem with Opencode but Hermes itself, I'm having the same problem but when I manually call the API the response is fine. I think Hermes is using incorrect headers or auth when calling the models.

curl https://opencode.ai/zen/go/v1/messages \     
  -H "x-api-key: ---" \           
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-m2.7",
    "max_tokens": 50,
    "messages": [
      {"role": "user", "content": "Say hello briefly"}
    ]
  }'
{"id":"","type":"message","role":"assistant","model":"MiniMax-M2.7","content":[{"thinking":"The user wants me to say hello briefly.","signature":"","type":"thinking"},{"text":"Hello! 👋","type":"text"}],"usage":{"input_tokens":44,"output_tokens":17},"stop_reason":"end_turn","base_resp":{"status_code":0,"status_msg":""},"cost":"0"}%

[jacob3712]

Closing this — after further investigation the bug is not on the OpenCode Go side at all.

Manual curl requests to the OpenCode Go API with the correct model name (e.g. minimax-m2.7) work fine. The HTTP 401 "Model not supported" error was being caused by a bug in the Hermes Agent CLI, which was silently replacing dots with hyphens in model names before making the API request (so minimax-m2.7 was being sent as minimax-m2-7).

I've filed the actual bug report here: NousResearch/hermes-agent#5211

Apologies for the noise!