Kimi
Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for agentic capabilities.
Kimi model releasesRank #181 across 526
Rank #196 across 436
Rank #207 across 357
Percentile score by analysis domain.
* Cost is inverted: lower input, output, and blended prices rank higher.
Higher bars mean stronger relative placement.
| Metric | Domain | Value | Rank |
|---|---|---|---|
| Artificial Analysis Intelligence Index | overall | 26.3 | #181 |
| Artificial Analysis Coding Index | coding | 22.1 | #196 |
| Artificial Analysis Math Index | math | 57.0 | #124 |
| MMLU-Pro | reasoning | 82.4% | #68 |
| reasoning |
| 76.6% |
| #155 |
| Humanity's Last Exam | reasoning | 7.0% | #228 |
| LiveCodeBench | coding | 55.6% | #130 |
| SciCode | coding, reasoning | 34.5% | #215 |
| MATH-500 | math | 97.1% | #26 |
| AIME | math | 69.3% | #41 |
| Output Speed | speed | 24.3 tok/s | #296 |
| Time to First Token | speed | 1.50s | #198 |
| Blended Price | cost | $1.04/M | #207 |
| Input Price | cost | $0.585/M | #205 |
| Output Price | cost | $2.40/M | #198 |
| Value Index | cost, overall | 25.3 | #195 |