[Bug] Zen API: `muse-spark-1.2-contributor-free` returns HTTP 500 for every request via API key, while other free models work

Summary

The model muse-spark-1.2-contributor-free is advertised in GET https://opencode.ai/zen/v1/models, but every POST /v1/chat/completions request for it returns:

{"type":"error","error":{"type":"error","message":"Internal server error"}}

with HTTP 500. This is 100% reproducible (15+ attempts over ~2 hours).

Other free models on the same key, at the same moment, return HTTP 200 normally. The same model works in the OpenCode desktop app under the same account, so the failure appears specific to the Zen API key path for this one model.

Environment

• Endpoint: https://opencode.ai/zen/v1
• Auth: Zen API key (Authorization: Bearer <key>)
• Account: logged in via GitHub
• Client: curl from a Linux container (also reproduced through Hermes Agent's OpenAI-compatible client)

Reproduction

curl -s -w "\nHTTP=%{http_code}" https://opencode.ai/zen/v1/chat/completions \
  -H "Authorization: Bearer $OPENCODE_ZEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"muse-spark-1.2-contributor-free","messages":[{"role":"user","content":"dis PONG"}],"max_tokens":40}'

Result: HTTP=500, {"type":"error","error":{"type":"error","message":"Internal server error"}}

Control tests (same key, same session)

Request	Result
hy3-free	200 OK
laguna-s-2.1-free	200 OK
nemotron-3-ultra-free	200 OK
mimo-v2.5-free	200, then FreeUsageLimitError (expected quota behaviour)
claude-haiku-4-5 (paid)	401 CreditsError — "No payment method" (clean, correct error)
muse-spark-1.2 (paid)	401 CreditsError (clean, correct error)
muse-spark-1.2-contributor-free	500 generic, always

The paid-model responses show the API returns precise, typed errors when a refusal is legitimate (billing). The contributor-free model instead returns an untyped 500, which looks like an unhandled server-side exception rather than an intentional refusal.

Variations tried — all return the same 500

• with max_tokens: 20, max_tokens: 2000, and with no max_tokens field
• with "stream": true
• with "variant": "xhigh" (the variant the desktop app records for this model)
• with "reasoning_effort": "xhigh"
• endpoint POST /v1/messages with x-api-key + anthropic-version headers
• endpoint POST /v1/messages with Authorization: Bearer
• model id opencode/muse-spark-1.2-contributor-free (prefixed)

Note: muse-spark-1.2-contributor-free:xhigh returns a different, correctly-typed error — {"type":"ModelError","message":"Model ... is not supported"} — confirming model-id validation happens before the crash, and that the bare id is the recognised one.

Note: POST /v1/messages returns 500 for every model tested (including hy3-free), so that endpoint appears unrelated / not supported.

Desktop app comparison

In the OpenCode desktop app, the same account can select and use this model. The app stores its selection as:

{"providerID":"opencode","modelID":"muse-spark-1.2-contributor-free","variant":"xhigh"}

So the model id matches exactly what the API rejects. The difference is the authentication path: desktop session (GitHub OAuth) vs Zen API key.

Expected behaviour

Either the model serves requests via the Zen API key like the other -free models, or it returns a typed, documented error explaining the requirement (e.g. an enrolment or consent flag). Note there is no visible "contributor programme" opt-in in the account settings, so if one is required it is not discoverable.

Impact

The model cannot be used by any API client (agents, SDKs, self-hosted tools) — only inside the desktop app.