You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
feat: document which API model (and price) each Kimi Code plan model id corresponds to #4781
We run Kimi Code CLI next to other coding agents and record each turn's token usage from wire.jsonl, so a team can compare the cost of the same work across models and providers. For accounts signed in with /login (the coding plan), we can't turn those tokens into an API-equivalent price. Tested on 2.3.1, macOS arm64.
What we see:
wire.jsonl and the model catalog carry the plan's model ids (kimi-for-coding, kimi-for-coding-highspeed, k3, k3-256k, k3.1-pro, k3.1-pro-highspeed), not the API platform's ids (kimi-k3.1-pro, kimi-k3, kimi-k2.7-code, …).
The API pricing page (platform.kimi.ai/docs/pricing/chat) lists prices only by API id.
The display names suggest a mapping (for example k3.1-pro → kimi-k3.1-pro), but we found it documented nowhere. kimi-for-coding (K2.8 Preview) has no API price at all. We'd rather not guess, because a wrong mapping gives silently wrong numbers.
Ask:
Document the mapping: for each coding-plan model id, the API model it corresponds to (or "no API equivalent"), on the Kimi Code docs or the pricing page.
Or expose it in the model catalog: an optional field on each plan model entry, for example "api_model": "kimi-k3.1-pro", next to the id that wire.jsonl already records.
Either one would let tools show an API-equivalent cost for plan usage, label it as such, and keep it correct when models change. A smaller, related request: publishing the 5-hour window and the monthly pool per plan in the same units would let tools turn "% used" into an expected amount of work.
Related: #2106 (session cost in the footer, closed).
Activity