| Benchmark | Claude 3.7 Sonnet | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 37 | 1.6 |
| AA Intelligence | 27.6 | 15.4 |
| AA-LCR | 62.3 | 41.3 |
| AA-Omniscience | -0.7 | -37.6 |
| aider_polyglot | 64.9 | 55.1 |
| AIME 2025 | 54.8 | 51.3 |
| AIR-Bench 2024 | 81.8 | 40.8 |
| anthropic_red_team | 99.7 | 97.1 |
| Artificial Analysis Coding Index | 36.4 | 23 |
| bbq | 92.1 | 96.7 |
| critpt | 0.9 | 0 |
| gdpval | 27.4 | 0 |
| GPQA Diamond | 84.8 | 68.4 |
| harmbench | 84.3 | 49.7 |
| HLE | 9.7 | 5.2 |
| hmmt_feb_2025 | 31.7 | 29.2 |
| IFBench | 48.3 | 41 |
| ifeval | 93.2 | 87.3 |
| math_500_em | 96.2 | 94 |
| mmmlu | 86.1 | 79.4 |
| multichallenge | 51.6 | 31.4 |
| OmniScience Accuracy | 28.2 | 25.4 |
| OmniScience Non-Hallucination | 60 | 23.3 |
| scicode | 40.3 | 35.8 |
| simple_safety_tests | 100 | 95.3 |
| simplebench | 46.4 | 27.2 |
| SWE-bench Verified | 70.3 | 42 |
| Terminal-Bench Hard | 21.2 | 15.2 |
| xstest | 96.4 | 97.1 |
| τ²-Bench Telecom (AA run) | 54.7 | 47.1 |
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.