| Benchmark | Claude 3.7 Sonnet | Kimi K2.5 |
|---|---|---|
| AA Agentic Index | 37 | 52.8 |
| AA Intelligence | 27.6 | 36 |
| AA-LCR | 62.3 | 73 |
| AA-Omniscience | -0.7 | -7.3 |
| AIME 2025 | 54.8 | 96.1 |
| arc_agi_1 | 13.6 | 65.3 |
| ARC-AGI-2 | 0 | 11.8 |
| arena_vision | 1195 | 1247 |
| Artificial Analysis Coding Index | 36.4 | 46.8 |
| critpt | 0.9 | 3.1 |
| Fortress | 38 | 41.1 |
| gdpval | 27.4 | 38.3 |
| GPQA Diamond | 84.8 | 87.9 |
| HLE | 9.7 | 50.2 |
| hmmt_feb_2025 | 31.7 | 95.4 |
| IFBench | 48.3 | 70.2 |
| MMMU-Pro | 60.1 | 78.5 |
| multichallenge | 51.6 | 61.4 |
| MultiNRC | 27.8 | 35.2 |
| OmniScience Accuracy | 28.2 | 35.2 |
| OmniScience Non-Hallucination | 60 | 50 |
| scicode | 40.3 | 49 |
| SEAL VISTA | 48.2 | 41.9 |
| simplebench | 46.4 | 46.8 |
| swe_bench_bash | 52.8 | 70.8 |
| SWE-bench Verified | 70.3 | 76.8 |
| Terminal-Bench Hard | 21.2 | 34.8 |
| τ²-Bench Telecom (AA run) | 54.7 | 95.9 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.