| Benchmark | Claude 4 Opus | Kimi K2.5 |
|---|---|---|
| AA Intelligence | 31.7 | 36 |
| AA-LCR | 40 | 73 |
| AIME 2025 | 75.5 | 96.1 |
| arc_agi_1 | 27 | 65.3 |
| ARC-AGI-2 | 8.6 | 11.8 |
| arena_vision | 1207 | 1247 |
| Artificial Analysis Coding Index | 34 | 46.8 |
| Fortress | 27.6 | 41.1 |
| frontiermath_tier_4 | 4.2 | 4.2 |
| GPQA Diamond | 79.6 | 87.9 |
| HLE | 12.3 | 50.2 |
| HMMT 2025 | 15.9 | 95.4 |
| hmmt_feb_2025 | 60 | 95.4 |
| IFBench | 53.7 | 70.2 |
| livebench | 74.6 | 69.1 |
| LiveCodeBench v6 | 47.4 | 85 |
| longbench_v2 | 55.6 | 61 |
| MMLU-Pro | 86.6 | 87.1 |
| multichallenge | 58.6 | 61.4 |
| MultiNRC | 33.9 | 35.2 |
| scicode | 40.9 | 49 |
| SEAL VISTA | 47 | 41.9 |
| simplebench | 58.8 | 46.8 |
| swe_bench_bash | 67.6 | 70.8 |
| SWE-bench Verified | 79.4 | 76.8 |
| Terminal-Bench Hard | 31.1 | 34.8 |
| vectara_answer_rate | 91 | 92.2 |
| vectara_avg_summary_length | 123.2 | 112 |
| vectara_factual_consistency | 88 | 85.8 |
| vectara_hallucination_rate ↓ | 12 | 14.2 |
| τ²-Bench | 57 | 85.4 |
| τ²-Bench Telecom (AA run) | 73.4 | 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.