Chat AI from China's Moonshot AI. Strong at long documents, and its K2-series open models draw global attention for agentic and coding performance.
Models
Pricing
- FreeFree
Web and app chat are essentially free.
- APIPay-as-you-go (cheap)
K2 models via low-cost APIs, widely adopted for coding.
※ For reference as of January 2026. Prices and plans change often — always check the official site for the latest.
Key features
Very long documents
Its trademark: digesting huge PDFs, papers, and books.
K2 Thinking (reasoning)
Reasoning mode of the K2 open models, noted globally for agentic strength.
OK Computer (agent)
An agent mode that goes from research to finished slides or web pages.
Research
Digs through the web and reports back with citations.
Tier lists featuring this service
Categories aggregate their whole subtree; threads count their own votes. Tap a row to open the tier list.
Not ranked in any tier list yet. Cast the first vote in a thread you care about.
Arena benchmark standings
arena.ai (formerly LMArena) blind-battle results converted into tiers for reference. See each category page for the conversion rules
- AText chatArena #16 ・ Elo 1475Representative model: kimi-k3-max
- AText chat (Japanese)Arena #15 ・ Elo 1475Representative model: kimi-k3-max
- ACodingArena #8 ・ Elo 1512Representative model: kimi-k3-max
- BCreative writingArena #25 ・ Elo 1449Representative model: kimi-k3-max
- SWeb development (Code Arena)Arena #2 ・ Elo 1674Representative model: kimi-k3-max
- CMathArena #13 ・ Elo 1488Representative model: kimi-k3-max
- SAgentArena #5 ・ +10.4%Representative model: Kimi K3 (Max)