fix(kimi-code): update known models metadata with current platform lineup - #10713
Conversation
…neup The /v1/models catalogue now serves kimi-for-coding, kimi-for-coding-highspeed, k3, and k3-256k. The metadata list is the fallback shown when the live fetch fails (e.g. expired token), and it still listed only kimi-for-coding, hiding k3 from the model picker. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5fbdc1557c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| pub const KIMI_CODE_KNOWN_MODELS: &[&str] = &[ | ||
| "kimi-for-coding", | ||
| "kimi-for-coding-highspeed", | ||
| "k3", |
There was a problem hiding this comment.
When a user selects the newly advertised k3 model, ProviderMetadata::new resolves it under provider name kimi_code, but map_provider_name does not translate that to the bundled catalog's kimi-for-coding/k3 entry. The resulting ModelInfo therefore uses the generic 128K context limit, which ProviderRegistry::normalize_model_config backfills into sessions, causing compaction at 128K even though the catalog declares a 1,048,576-token context. Add the provider-name mapping or construct these entries with their explicit limits.
Useful? React with 👍 / 👎.
2 participants
Summary
kimi_code's/v1/modelscatalogue now serveskimi-for-coding,kimi-for-coding-highspeed,k3, andk3-256k(verified live againstapi.kimi.com/coding/v1/models)KIMI_CODE_KNOWN_MODELSstill listed onlykimi-for-coding; it's the fallback shown when the live fetch fails (e.g. expired token), so k3 was hidden from the model pickerk3/k3-256k, notkimi-k3Test plan
cargo test -p goose --lib kimicode(12 pass)cargo clippy -p goose --lib -- -D warnings