🤖 fix: route Copilot gpt-5.4-mini through a supported API mode

[alecsg77]

GitHub Copilot gpt-5.4-mini is routed to unsupported Chat Completions endpoint

Environment

• Mux: v0.28.1-nightly.12-7-gd25b50ecc-dirty (d25b50ecc)
• Platform: Linux x86_64
• Provider/model: github-copilot:gpt-5.4-mini
• Related work: open PR 🤖 fix: adapt Copilot Responses streams for AI SDK #2967

Reproduction

mux run --runtime local --mode exec --thinking low --json --hide-costs \
  --model github-copilot:gpt-5.4-mini \
  'Respond with exactly OK. Do not call tools. Do not add punctuation or explanation.'

Actual result

Mux selects routeProvider: "github-copilot" and posts to:

https://api.githubcopilot.com/chat/completions

Copilot returns HTTP 400:

model "gpt-5.4-mini" is not accessible via the /chat/completions endpoint
code: unsupported_api_for_model

The refreshed Copilot /models catalog contains gpt-5.4-mini. Control models github-copilot:gpt-5.4 and github-copilot:gpt-5-mini passed the same smoke test and a bash tool test.

Expected result

Mux should either:

1. route this provider/model pair through a supported API mode; or
2. mark/hide the model as unavailable until a supported transport exists.

The evidence proves that Chat Completions is rejected. It does not by itself prove that simply switching to Responses is sufficient.

Suspected area

src/common/utils/copilot/modelRouting.ts currently selects Responses only for model IDs containing -codex; all other Copilot models use Chat Completions. Dispatch occurs in src/node/services/providerModelFactory.ts.

---

Generated with mux • Model: openai:gpt-5.6-terra • Thinking: high • Cost: $1.34

[linear-code]

MUX-64