Kimi

★ 4/5 · AI Agent Platform

Kimi is Moonshot AI’s platform for agentic programming and knowledge work, with tools for research, documents, presentations, websites, spreadsheets and design.

Pros

Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with 104B active parameters, native vision and a 1,048,576-token context window, and Moonshot published its weights openly in July 2026, so the same frontier model can be self-hosted or bought as an API
the API stays cheap for its class at $3.00/M input and $15.00/M output with cache hits at $0.30/M, while the coding-tuned kimi-k2.7-code runs $0.95/M input and $4.00/M output over a 262,144-token window and offers a high-speed variant at roughly 180 tokens per second. The kimi.com side is a genuine work platform rather than a chat box, bundling Agent Swarm, Deep Research, Docs, Sheets and Slides, scheduled tasks, one-click website deployment with a database, 15+ data plugins and the Kimi Code and Kimi Claw agents, and the free Adagio tier costs nothing while still allowing agent tasks, two projects and full plugin access.

Cons

Frontier pricing has caught up with the West: K3's $15.00/M output sits in the same band as the GPT and Claude flagships, so the cheap-Chinese-model framing no longer applies at the top of the range, and K3 is excluded from the Batch API's 40% discount that K2.6 and K2.7 Code still receive. Moonshot retires models hard, with kimi-k2.5 and the entire moonshot-v1 series switched off on 31 August 2026 and now returning 404s, so pinned integrations break on a schedule. On the consumer side the membership draws on a single shared credit pool metered by token usage across chat, agents, Kimi Code and image generation, which makes spend difficult to forecast
the 1M-token long-chat capacity is gated to the $99 Allegro tier and above
free Adagio allows only one concurrent agent task, two projects and 500MB of storage with no Agent Swarm, Kimi Code or Kimi Claw
and at the time of writing every paid tier displays a Join Waitlist button while Moonshot splits Kimi and Kimi Code into separate plans. Data residency under a Chinese operator remains a blocker for regulated buyers.

Cost

API: kimi-k3 $3.00/M input, $0.30/M cache-hit input, $15.00/M output at a 1,048,576-token context; kimi-k2.7-code $0.95/M input, $0.19/M cached, $4.00/M output at 262,144 tokens, with the high-speed variant at double ($1.90/M input, $8.00/M output); kimi-k2.6 $0.95/M input, $0.16/M cached, $4.00/M output. The Batch API bills 60% of standard rates on K2.6 and K2.7 Code but does not cover K3. Consumer memberships are Adagio free, Moderato $19/month ($15/month billed annually, $180/year), Allegretto $39 ($31, $372/year), Allegro $99 ($79, $948/year) and Vivace $199 ($159, $1,908/year); prices exclude tax, and Moonshot has announced a repricing that separates Kimi and Kimi Code benefits, so paid tiers currently sit behind a waitlist.

Verdict

Best for teams doing long-context agentic work such as deep research, document and deck production or coding against a 262K window, who also want a frontier open-weight model they can self-host; less suited to anyone who needs stable model IDs, predictable per-seat billing or a Western data-residency guarantee.

Visit Kimi