Need a China coding plan (GLM / MiniMax / Kimi / Volcengine / Xiaomi)? Email me at [email protected]
Moonshot AI · Kimi Code
Kimi K3 (2.8T parameters, 1M context, up to 260 tokens/s) leads the lineup, and the new membership system is live: Free / Go ¥49 / Plus ¥99 / Pro ¥199 / Max ¥699. Weekly quota limits are gone - only a rolling 5-hour window remains - Go has no coding quota at all, and Kimi Code starts at Plus. Existing subscribers keep their old plans.
Kimi K3 (k3 / k3-256k) plus the kimi-for-coding family: K2.8 Preview and the K2.7 Code high-speed build.
Moonshot’s flagship: 2.8 trillion parameters, up to 1M-token context with low/high/max thinking levels, up to 260 tokens/s. The k3-256k variant gives identical results within 256K at roughly half the consumption (no video input). Kimi Code and K3 unlock from Plus; full 1M context and the high-speed build need Pro.
The workhorse for code completion and routine development, with context up to 1M. Served over both the OpenAI and Anthropic protocols; requires Kimi Code, so Plus or above.
K2.7 Code high-speed build (~5-6x faster) for low-latency loops; unlocks from Pro.
Free to Max at ¥49-¥699 per month (annual billing saves up to ¥1,680). Kimi Code needs Plus or above — Go has no coding quota. Weekly limits were removed for new members; quotas now use a rolling 5-hour window within a monthly cap.
Free
FreeCasual users and newcomers
Go
¥39/mo annually (¥468/yr)
Individual developers who mainly use Kimi chat
Plus
Recommended · Kimi Code¥79/mo annually (¥948/yr)
Daily high-frequency coders
Pro
1M context · high speed¥159/mo annually (¥1,908/yr)
Heavy agent workflows
Max
Top tier¥559/mo annually (¥6,708/yr) · saves ¥1,680
Power users and teams
Coding is only the start.
K3 / K2.7 / K2.8 for generation, refactoring and debugging.
Spin up site prototypes and landing pages.
Generate technical and API documentation.
Quickly build shareable presentations.
K3 is included from Plus (¥99/mo), together with the K3-256K budget variant; Go and Free have no Kimi Code quota at all. K3 with 1M-token context and the high-speed coding model need Pro (¥199/mo) or Max.
A model ID (k3-256k) that returns identical results to K3 within 256K context while consuming roughly half the quota. It does not support video input. Use it for daily and single-file tasks to save quota; the full 1M-context K3 consumes about 2x.
The new plans are live at unchanged pricing: Free ¥0, Go ¥49, Plus ¥99, Pro ¥199, Max ¥699 (annual billing saves up to ¥1,680). Weekly quota limits are removed for new members — only a rolling 5-hour window remains, inside a monthly cap. Kimi Code starts at Plus, and existing subscribers keep their old plans (Adagio / Andante / Moderato / Allegretto / Allegro map to Free / Go / Plus / Pro / Max).
New memberships no longer refresh on a weekly cycle: quotas use a rolling 5-hour frequency window against a monthly total, and the Kimi Code quota freezes once the monthly cap is reached. All devices and API keys share the same quota; unused quota does not roll over.
Extra Usage is a pay-as-you-go balance that covers usage after the subscription quota is exhausted. Top-ups start at ¥25 each, up to 10 per day with a daily cumulative ¥3,000, and a ¥10,000 balance cap; the balance never expires, can be stacked, and is shared with Kimi on the web.
Available from Plus: Pro gets 40 cluster runs/mo with 4 subtasks; Max gets 8 subtasks and 4 parallel Agents.
Kimi CLI, Kimi Code for VS Code, Claude Code and Roo Code, all sharing the same plan quota. The API speaks both OpenAI and Anthropic protocols — api.kimi.ai/coding overseas and api.kimi.com/coding in China.