| Benchmark | Claude 3.7 Sonnet | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 37 | 49.7 |
| AA Intelligence | 27.6 | 38.7 |
| AA-LCR | 62.3 | 73 |
| AA-Omniscience | -0.7 | 10.1 |
| AIME 2025 | 54.8 | 99.7 |
| arc_agi_1 | 13.6 | 84.7 |
| ARC-AGI-2 | 0 | 33.6 |
| arena_vision | 1195 | 1260 |
| Artificial Analysis Coding Index | 36.4 | 42.6 |
| critpt | 0.9 | 8.6 |
| gdpval | 27.4 | 35.2 |
| GPQA Diamond | 84.8 | 90.4 |
| harmbench | 84.3 | 70 |
| HLE | 9.7 | 43.5 |
| hmmt_feb_2025 | 31.7 | 97.5 |
| IFBench | 48.3 | 78 |
| mmmlu | 86.1 | 91.8 |
| MMMU-Pro | 60.1 | 81.2 |
| OmniScience Accuracy | 28.2 | 53.4 |
| OmniScience Non-Hallucination | 60 | 7.6 |
| scicode | 40.3 | 50.6 |
| simplebench | 46.4 | 61.1 |
| swe_bench_bash | 52.8 | 75.8 |
| SWE-bench Verified | 70.3 | 78 |
| Terminal-Bench Hard | 21.2 | 38.6 |
| τ²-Bench Telecom (AA run) | 54.7 | 80.4 |
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.