[Zen] muse-spark-1.3-contributor-free returns HTTP 500 via API key, while 1.2-contributor-free works

[qian-123456]

Summary

Calls to muse-spark-1.3-contributor-free through the Zen API (https://opencode.ai/zen/v1/chat/completions) with a workspace API key consistently return HTTP 500 Internal server error, while the identical request to muse-spark-1.2-contributor-free succeeds. The model itself is healthy (serves fine via the desktop/OAuth path), so this looks like a broken upstream routing/integration for the 1.3 free SKU on the API-key path.

Repro (minimal, client-independent)

curl https://opencode.ai/zen/v1/chat/completions \
  -H "Authorization: Bearer $OPENCODE_ZEN_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"muse-spark-1.3-contributor-free","messages":[{"role":"user","content":"hi"}]}'
# -> HTTP 500 {"type":"error","error":{"type":"error","message":"Internal server error"}}

Same request with "model":"muse-spark-1.2-contributor-free" succeeds (verified end-to-end in a third-party client, full reply received).

Controlled comparison (same session, same key, same endpoint, prompt = hi, 2026-09-04 ~01:51-01:53 UTC)

model	result
muse-spark-1.2-contributor-free	HTTP 200, normal reply
muse-spark-1.3-contributor-free	HTTP 500 x4 (initial + 3 client retries, budget exhausted)

Also reproduced with stream: true and with/without max_tokens — 1.3-free 500s in every shape; 1.2-free works through the full client.

Ruled out on my side

• Workspace API key is valid: GET /zen/v1/models returns 200 with 66 models, including both spark free SKUs (identical metadata, same created timestamp).
• Both Muse Spark 1.2 Free and Muse Spark 1.3 Free toggles are enabled in the workspace model settings.
• Other SKUs return clean, structured errors through the same key/path (e.g. paid muse-spark-1.2 -> CreditsError: No payment method; mimo-v2.5-free -> FreeUsageLimitError rate limit; deepseek-v4-flash-free -> Model is unavailable). Only the spark-free SKUs return bare 500s, and only 1.3 is unusable end-to-end.
• GET https://opencode.ai/ returns 200 from the same network (no IP/CF ban).

Environment

• Workspace: wrk_01KXYHCRYNRDJ0KDZ49JCGCRPN (no billing method attached; free SKUs only)
• Observed via direct curl + Oh My Pi 18.0.4 (openai-completions API, error logged as errorStatus: 500, errorMessage: 500 Internal server error)
• First observed 2026-09-03, still reproducing 2026-09-04

Expected

muse-spark-1.3-contributor-free serves requests over the Zen API like its 1.2 counterpart, or returns a structured error explaining why it cannot.

Happy to provide request IDs/timestamps or test a fix against the same workspace key.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [Bug] Zen API: muse-spark-1.2-contributor-free returns HTTP 500 for every request via API key, while other free models work #45744: Same symptom — muse-spark-1.2-contributor-free returns HTTP 500 via Zen API key while the desktop/OAuth path works; the pattern of a contributor-free SKU failing specifically on the API-key path is identical.
• OpenCode Zen muse-spark-1.2-contributor-free suddenly returns HTTP 500 with OpenAI-compatible provider #44847: Same error (muse-spark-1.2-contributor-free HTTP 500 via OpenAI-compatible provider on Zen).
• muse-spark-1.2-contributor-free returns HTTP 500 Internal server error #45584: Same model, same 500 error on the Zen API.