| Benchmark | GPT-4o | Kimi K2.6 |
|---|---|---|
| AA Agentic Index | 9.7 | 58.7 |
| AA Intelligence | 11.2 | 45.1 |
| AA-LCR | 55 | 76.7 |
| AA-Omniscience | -10.5 | 6.4 |
| arena_vision | 1162 | 1264 |
| Artificial Analysis Coding Index | 24.2 | 61.8 |
| Chinese SimpleQA (C-SimpleQA) | 64.6 | 75.9 |
| critpt | 0 | 8 |
| gdpval | 0 | 41.4 |
| GPQA Diamond | 70.1 | 91.1 |
| HLE | 5.3 | 37.5 |
| IFBench | 36 | 76 |
| LiveCodeBench | 38.3 | 89.6 |
| LiveCodeBench v6 | 30.9 | 89.6 |
| mathvision | 30.4 | 93.2 |
| MMLU-Pro | 74.7 | 87.1 |
| MMMU-Pro | 59.9 | 80.1 |
| OmniScience Accuracy | 23.7 | 32.8 |
| OmniScience Non-Hallucination | 62.1 | 60.7 |
| OSWorld-Verified | 5 | 73.1 |
| scicode | 33.4 | 53.5 |
| SWE-bench Verified | 38.8 | 80.2 |
| Terminal-Bench Hard | 8.3 | 43.9 |
| vectara_answer_rate | 93.8 | 99.7 |
| vectara_avg_summary_length | 86.6 | 116.7 |
| vectara_factual_consistency | 90.4 | 89.2 |
| vectara_hallucination_rate ↓ | 9.6 | 10.8 |
| τ²-Bench Telecom (AA run) | 28.9 | 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.