Kimi K3 docs added: Moonshot’s open-weight frontier coding model for long-running agent work; billed from the third-party Other Models pool with flat per-token pricing (no long-context surcharge).
Changelog
Documentation updates
Models
Kimi K3 docs added: Moonshot’s open-weight frontier coding model for long-running agent work; billed from the third-party Other Models pool with flat per-token pricing (no long-context surcharge).
Kimi K2.7 Code: Kimi K3 recommended for new work; K2.7 Code stays available as a lower-cost option.
Grok 4.5: Documents high/medium/low effort levels (default high). Removed the doubled included-usage promo through July 21, 2026; standard and Fast on-demand rates unchanged.
API & SDK
Cloud Agents API: Clarified that the API and SDKs run Cursor agent workflows (not standalone model inference or chat completions); Auto / auto-smart uses Cursor Router.
Cursor Router (TypeScript SDK): Use model id auto-smart with optimize_for (cost, balanced, or intelligence); call Cursor.models.list() to confirm availability. Teams/Enterprise; Enterprise admins must enable Router.
TypeScript SDK: Expanded Router docs (selection, catalog discovery, per-run overrides). Removed cloud agent metadata docs. Package version pin dropped from the types note.
Python SDK: Defaults updated — local.force defaults to None (set True to expire a stuck run); mode and cloud env default to None (server uses agent mode and Cursor-hosted VMs when omitted).
