- Declare the real ~1M contextLength/maxOutputTokens on the muse-spark-1.2 /
muse-spark-1.2-contributor-free registry entries (opencode + opencode-zen)
instead of silently falling back to the 200000 provider default (#12681).
- Send x-api-key instead of Authorization: Bearer for the openai-responses
format on the main OpenCode Zen host, fixing a 401 on Muse Spark
Contributor's /v1/responses route; scoped by baseUrl so opencode-go (a
different upstream) keeps Bearer (#12633).