| Benchmark | Kimi K2.5 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 36 | 10 |
| AA-LCR | 73 | 20.3 |
| AA-Omniscience | -7.3 | -52.2 |
| Artificial Analysis Coding Index | 46.8 | 11.9 |
| critpt | 3.1 | 0 |
| Fortress | 41.1 | 56.4 |
| GPQA Diamond | 87.9 | 49.1 |
| HLE | 50.2 | 4.2 |
| IFBench | 70.2 | 36.9 |
| livebench | 69.1 | 52.3 |
| longbench_v2 | 61 | 39.4 |
| MMLU-Pro | 87.1 | 71.6 |
| OmniScience Accuracy | 35.2 | 17.6 |
| OmniScience Non-Hallucination | 50 | 15.3 |
| scicode | 49 | 26.7 |
| SWE-bench Verified | 76.8 | 23.8 |
| Terminal-Bench Hard | 34.8 | 4.5 |
| τ²-Bench Telecom (AA run) | 95.9 | 34.5 |
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.