| Benchmark | Claude 3.7 Sonnet | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 37 | 18.8 |
| AA Intelligence | 27.6 | 17.7 |
| AA-LCR | 62.3 | 65.7 |
| AA-Omniscience | -0.7 | -29.8 |
| aider_polyglot | 64.9 | 61.9 |
| AIME 2025 | 54.8 | 72 |
| AIR-Bench 2024 | 81.8 | 68.7 |
| anthropic_red_team | 99.7 | 98.8 |
| arc_agi_1 | 13.6 | 33.3 |
| ARC-AGI-2 | 0 | 2.5 |
| arena_vision | 1195 | 1214 |
| Artificial Analysis Coding Index | 36.4 | 22.2 |
| bbq | 92.1 | 97.7 |
| critpt | 0.9 | 1.4 |
| gdpval | 27.4 | 11.9 |
| GPQA Diamond | 84.8 | 82.8 |
| harmbench | 84.3 | 62.6 |
| HLE | 9.7 | 12.1 |
| hmmt_feb_2025 | 31.7 | 64.2 |
| IFBench | 48.3 | 50.3 |
| MMMU | 75 | 79.7 |
| MMMU-Pro | 60.1 | 69.1 |
| OmniScience Accuracy | 28.2 | 26.1 |
| OmniScience Non-Hallucination | 60 | 24.7 |
| scicode | 40.3 | 35.9 |
| SEAL VISTA | 48.2 | 49.1 |
| simple_safety_tests | 100 | 98 |
| simplebench | 46.4 | 41.2 |
| swe_bench_bash | 52.8 | 28.7 |
| SWE-bench Verified | 70.3 | 60.4 |
| Terminal-Bench Hard | 21.2 | 13.6 |
| xstest | 96.4 | 98.8 |
| τ²-Bench Telecom (AA run) | 54.7 | 31.6 |
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.