| Benchmark | Claude 3.7 Sonnet | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 37 | 6.3 |
| AA Intelligence | 27.6 | 8.5 |
| AA-LCR | 62.3 | 24.7 |
| AA-Omniscience | -0.7 | -17.1 |
| aider_polyglot | 64.9 | 5.8 |
| AIR-Bench 2024 | 81.8 | 58.6 |
| anthropic_red_team | 99.7 | 96.5 |
| Artificial Analysis Coding Index | 36.4 | 14.5 |
| bbq | 92.1 | 94.5 |
| critpt | 0.9 | 0 |
| Fortress | 38 | 20.6 |
| gdpval | 27.4 | 0 |
| GPQA Diamond | 84.8 | 51.5 |
| harmbench | 84.3 | 62.7 |
| HLE | 9.7 | 4.2 |
| IFBench | 48.3 | 39 |
| ifeval | 93.2 | 88.6 |
| math_500_em | 96.2 | 73.8 |
| mmmlu | 86.1 | 73.8 |
| OmniScience Accuracy | 28.2 | 23.2 |
| OmniScience Non-Hallucination | 60 | 49 |
| scicode | 40.3 | 29.9 |
| simple_safety_tests | 100 | 98.8 |
| simplebench | 46.4 | 23 |
| SWE-bench Verified | 70.3 | 24.5 |
| Terminal-Bench Hard | 21.2 | 6.8 |
| xstest | 96.4 | 95.9 |
| τ²-Bench Telecom (AA run) | 54.7 | 19 |
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.