Kimi K2 (Thinking): benchmark results
Kimi K2 evaluated with thinking enabled. Provider: Moonshot. Released 2025-11-06. Access: Open.
Aggregate Rating 1645 ± 4, rank #537 of 2168 rated models, from 786 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CLEM Wordle | 73 | Game Clemscore (%), text v3.0; published played × quality / | 100 |
| LingxiDiagBench (Dynamic, Free-Form) - 4-Class Accuracy | 36 | Accuracy (%, 0-100) on four-way classification (pure depress | 100 |
| LogiScope-VQA - Warehouse Commonsense | 96 | Micro-averaged accuracy (%) | 100 |
| SuperCLUE General (September 2025) - Science Reasoning | 54.31 | Score | 98.6 |
| BenchTable - Utility | 82.6 | Weighted Score (%) | 98.1 |
| SuperCLUE General (September 2025) - Math Reasoning | 80.91 | Score | 97.2 |
| ReLE - Agents - BFCL-V3 | 73.5 | Accuracy (%) | 96.3 |
| AA LiveCodeBench | 85.29 | Pass@1 (%) | 96.2 |
| CLEM TextMapWorld GraphReasoning | 84.28 | Game Clemscore (%), text v3.0; published played × quality / | 96.2 |
| AA AIME 2025 | 94.67 | Accuracy (%) | 96.1 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 46 | Accuracy (%) | 96.1 |
| RewardBench 2 Math | 89.62 | Score (%) | 95.8 |
Interactive version: theaggregate.ai/model?slug=kimi-k2-thinking · How It Works · Data refreshed daily, snapshot 2026-10-20.