Provider API: meta/muse-spark-1.3* fails with misleading "Stream error occurred" — non-streaming reveals HTTP 503 overloaded_error

Summary

All Muse Spark 1.2/1.3 requests to the Provider API (POST /provider/v1/chat/completions) fail on a GOAT-plan account. Streaming returns HTTP 200 + an SSE error chunk {"code":"error","message":"Stream error occurred"}; non-streaming returns HTTP 503 with the real reason: overloaded_error ("Upstream model provider is temporarily unavailable"). Streaming-only clients (e.g. the AI SDK / opencode) never see the real error — it surfaces as "Stream error occurred: error", which looks like a client bug.

Not an entitlement issue: meta/muse-spark-1.3-contributor has a $20/mo GOAT allowance, and meta/muse-spark-1.1 correctly returns MODEL_NOT_IN_PLAN (permission_error).

Expected Behavior

Muse Spark 1.3 streams work on GOAT. At minimum, when the upstream provider is unavailable, the SSE error chunk should carry the real error type/status/message instead of a generic "Stream error occurred".

Actual Behavior

• Streaming: HTTP 200, SSE:
  data: {"id":"gen_...","error":{"code":"error","message":"Stream error occurred"},"choices":[{"index":0,"delta":{"content":""},"finish_reason":"error"}]}
• Non-streaming: HTTP 503 {"error":{"message":"Upstream model provider is temporarily unavailable. Please try again in a moment.","type":"overloaded_error"}}
• Consistent across ~20 minutes of attempts; affected: meta/muse-spark-1.2, meta/muse-spark-1.2-contributor, meta/muse-spark-1.3, meta/muse-spark-1.3-contributor.
• Control: deepseek/deepseek-v4-flash works fine on the same key/endpoint.

Steps to reproduce

1. curl -N https://api.commandcode.ai/provider/v1/chat/completions -H "Authorization: Bearer $COMMANDCODE_API_KEY" -H "Content-Type: application/json" -d '{"model":"meta/muse-spark-1.3-contributor","messages":[{"role":"user","content":"hi"}],"stream":true,"max_tokens":32}'
2. Repeat with "stream":false → HTTP 503 overloaded_error.
3. Control: same request with deepseek/deepseek-v4-flash → HTTP 200, streams normally.

Command Code Version

N/A — reporting against the Provider API directly (client: opencode-cmd-provider 1.7.0, sends x-command-code-version: 1.15.1).

Operating System

Linux

Terminal/IDE

opencode TUI

Shell

zsh

Fix prompt (optional)

"Provider API /provider/v1/chat/completions returns a generic SSE error 'Stream error occurred' for Muse Spark 1.2/1.3 when the upstream provider is overloaded/unavailable (non-streaming correctly returns 503 overloaded_error). Propagate the real upstream error type/message/status in the SSE error chunk, and check why meta/muse-spark-1.3* fails upstream while deepseek works."

Additional context

• x-trace-id: 16ab2fb3cdabdb81f6f8df15f1a57a11 (streaming attempt, 2026-09-10 17:02:33 UTC)
• Account plan: GOAT
• Related: #809 (different surface — opencode/opencode-go BYOK), #799 (usage)