OpenCode Go: gpt-5.6-luna still returns HTTP 500 'Internal server error' (confirmed 2026-09-07)

Summary

gpt-5.6-luna on the OpenCode Go plan still returns HTTP 500 {"type":"error","error":{"type":"error","message":"Internal server error"}} for every request. Confirmed live on 2026-09-07 with a working Go subscription and the same API key that succeeds on other catalog models. This duplicates earlier reports (#47463, #43565) - still unfixed, adding a fresh data point with a control test.

Reproduction (2026-09-07, plan: OpenCode Go subscription)

curl -sS -o /dev/null -w "%{http_code}\n" \
  -H "Authorization: Bearer $OPENCODE_GO_API_KEY" \
  -H "x-opencode-session: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","messages":[{"role":"user","content":"Reply with exactly: pong"}],"max_tokens":64}' \
  https://opencode.ai/zen/go/v1/chat/completions

Result: 500 with body

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

Control test (same key, same session, same endpoint)

# model: deepseek-v4-pro

Result: 200 with normal completion (content: "pong", finish_reason: "stop").

The same single key succeeds on other models (e.g. deepseek-v4-pro), so this is not an auth or billing issue - the model is listed in the catalog but not wired on the Go router.

Expected behavior

gpt-5.6-luna either works on the Go plan like other catalog models, or is removed/marked unavailable in the catalog so subscribers cannot select a model that always fails.

Context

• Follow-up of #47463 (Bucket 1: gateway 500s) and #43565 (empty response/403) - both still open.
• Related service stability thread: #36889.