| Benchmark | Kimi K2.5 | o3 |
|---|---|---|
| AA Agentic Index | 52.8 | 36.1 |
| AA Intelligence | 36 | 31.1 |
| AA-LCR | 73 | 73.3 |
| AA-Omniscience | -7.3 | -15.3 |
| AIME 2025 | 96.1 | 89.2 |
| arc_agi_1 | 65.3 | 60.8 |
| ARC-AGI-2 | 11.8 | 6.5 |
| arena_vision | 1247 | 1217 |
| Artificial Analysis Coding Index | 46.8 | 38.4 |
| browsecomp | 74.9 | 49.7 |
| critpt | 3.1 | 1.1 |
| Fortress | 41.1 | 16 |
| frontiermath_tier_4 | 4.2 | 2.1 |
| gdpval | 38.3 | 12.8 |
| GPQA Diamond | 87.9 | 83.3 |
| HLE | 50.2 | 20.3 |
| hmmt_feb_2025 | 95.4 | 77.5 |
| IFBench | 70.2 | 71.4 |
| longbench_v2 | 61 | 58.8 |
| MathVista | 90.1 | 86.8 |
| MMLU-Pro | 87.1 | 85 |
| MMMU-Pro | 78.5 | 76.4 |
| multichallenge | 61.4 | 60.4 |
| MultiNRC | 35.2 | 45.5 |
| OmniScience Accuracy | 35.2 | 38.6 |
| OmniScience Non-Hallucination | 50 | 12.9 |
| scicode | 49 | 41 |
| simplebench | 46.8 | 53.1 |
| simpleqa_verified | 36.9 | 53 |
| strongreject | 99.5 | 97 |
| swe_bench_bash | 70.8 | 58.4 |
| SWE-bench Verified | 76.8 | 69.1 |
| Terminal-Bench Hard | 34.8 | 37.1 |
| VideoMMMU | 86.6 | 83.3 |
| τ²-Bench Telecom (AA run) | 95.9 | 80.7 |
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.