Problem

When adding new GitHub Copilot models via config (e.g., gpt-5.3-codex), they fail with "model not supported" error even though the model works in VS Code Copilot.

{
  "provider": {
    "github-copilot": {
      "models": {
        "gpt-5.3-codex": {
          "modalities": {
            "input": ["text", "image"],
            "output": ["text"]
          }
        }
      }
    }
  }
}

Root Cause

Two issues compound to cause this:

1. Wrong SDK package: Config-defined models fall back to @ai-sdk/openai-compatible which lacks the responses() method needed for /responses endpoint. Models like gpt-5.3-codex require this endpoint, not /chat/completions.

2. Plugin loading blocked: External auth plugins (like opencode-copilot-auth) that use the correct GitHub App client ID (Iv1.b507a08c87ecfe98) were explicitly skipped from loading in plugin/index.ts.

Technical Details

• The bundled @ai-sdk/github-copilot SDK in ./sdk/copilot/ has the responses() method
• The npm package @ai-sdk/openai-compatible does not
• The npm override in provider.ts ran BEFORE config processing, so config-defined models never got the correct SDK
• Line 56 in plugin/index.ts contained: if (plugin.includes("opencode-copilot-auth")) continue which prevented external plugins from loading

Related

• models.dev PR: feat(github-copilot): add gpt-5.3-codex model models.dev#857
• Discussion thread with affected users reporting same issue

Fix

• PR: fix(copilot): enable gpt-5.3-codex and other models requiring /responses endpoint #13485