Refresh MiniMax model suggestions for M3.1 Flash and M3
Reason: Refresh MiniMax suggestions to MiniMax-M3.1-Flash and MiniMax-M3, including the current routes available through OpenRouter and OpenCode.
- Default native global/China suggestions to Flash followed by M3 in both the generator and runtime-generated types.
- Filter ordinary MiniMax text suggestions in the known OpenRouter and OpenCode catalogs during generation and runtime refresh. Retain only current routes actually returned by the catalog; do not synthesize missing gateway IDs.
- Narrowly regenerate MiniMax entries while preserving unrelated providers, nontext/embedding models and explicit saved/custom model resolution, including legacy IDs.
- Test generation, runtime snapshots, absent current routes, custom providers, nontext preservation and explicit legacy/custom gateway resolution after refresh.
- Include compatible SDK captures for both native models and regions. Requests omit thinking, effort and service tier, and never inject reasoning_split=false, leaving supported API defaults in effect.
- Include the required core package changeset.
The exact gateway routes were verified in models.dev and the gateway catalogs on 2026-10-19:
- OpenRouter:
minimax/minimax-m3.1-flash:free,minimax/minimax-m3. - OpenCode:
minimax-m3.1-flash-free,minimax-m3.
Package validation (all passed): pinned pnpm dependency installation and internal/logger prerequisite builds; 18 focused registry/catalog/wire tests; full core vitest run --typecheck (112 files, 1,410 passed, 2 skipped, 1 todo, no type errors); core pnpm typecheck; core pnpm build; scoped Biome; git diff --check; diff secret scan.
Commitlint currently reports type-empty and subject-empty for the published gateway followup commit subject.
Dependencies and checks ran without deployment credentials. No paid inference was performed. Catalog pricing and token limits are not copied or inferred from route availability.
Model reference: https://platform.minimax.io/docs/api-reference/text-chat-openai.md