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
[BUG] "Model does not support this protocol" for gemini-4-argon-contributor #259
Sending a chat message with gemini-4-argon-contributor (OpenCode Go provider) fails immediately with a 400 error: "Model does not support this protocol." Every message to this model fails.
🔄 Steps to reproduce
Open Copilot Chat and select gemini-4-argon-contributor from the OpenCode Go models in the model picker.
Send any message.
Observe the error: OpenCode Go API request failed (400) model=gemini-4-argon-contributor payloadBytes=122820: Model does not support this protocol.
✅ What did you expect?
The request should stream a response. If the model cannot be used with this protocol, it should not appear in the model picker, or the error should say which protocol is expected.
🖥️ Environment
Extension version: 0.7.9
VS Code version: 1.141.1
Provider: OpenCode Go
Model: gemini-4-argon-contributor
📸 Logs or screenshots (optional)
OpenCode Go API request failed (400) model=gemini-4-argon-contributor payloadBytes=122820: Model does not support this protocol. OpenCodeRequestError: OpenCode Go API request failed (400) model=gemini-4-argon-contributor payloadBytes=122820: Model does not support this protocol. at buildOpenCodeRequestError (...\ltmoerdani.opencode-copilot-chat-0.7.9\out\errors.js:40:12) at streamOpenCodeResponse (...\out\transports\engine.js:320:73) at streamChatCompletions (...\out\transports\chatCompletions.js:26:9) at OpenCodeProvider.provideLanguageModelChatResponse (...\out\provider\OpenCodeProvider.js:388:13)
Root cause: the gemini entry in our model registry was restricted to the Zen vendor, so on Go the lookup fell through to the catch-all row and the request went to /chat/completions — while the gateway serves that model Google-native. The registry row now applies to every vendor, and Go's Google route points at the right base (/inference/go/v1/models, since /zen/go/v1 doesn't serve :streamGenerateContent at all).
This ships in v0.7.10, which should land on the Marketplace in the next few days — you'll get it automatically via extension update. Your diagnostics dump made this easy to pin down, appreciated.
Activity
ltmoerdani commentedon Oct 20, 2026
Fixed via #260 (thanks @MidhunCN for the PR).
Root cause: the
geminientry in our model registry was restricted to the Zen vendor, so on Go the lookup fell through to the catch-all row and the request went to/chat/completions— while the gateway serves that model Google-native. The registry row now applies to every vendor, and Go's Google route points at the right base (/inference/go/v1/models, since/zen/go/v1doesn't serve:streamGenerateContentat all).This ships in v0.7.10, which should land on the Marketplace in the next few days — you'll get it automatically via extension update. Your diagnostics dump made this easy to pin down, appreciated.