| Benchmark | Llama 3.1 Instruct 405B | o3 |
|---|---|---|
| AA Agentic Index | 6.3 | 36.1 |
| AA Intelligence | 8.5 | 31.1 |
| AA-LCR | 24.7 | 73.3 |
| AA-Omniscience | -17.1 | -15.3 |
| aider_polyglot | 5.8 | 81.3 |
| AIME 2024 | 23.3 | 91.6 |
| AIR-Bench 2024 | 58.6 | 84.5 |
| anthropic_red_team | 96.5 | 98.3 |
| Artificial Analysis Coding Index | 14.5 | 38.4 |
| bbq | 94.5 | 97.9 |
| critpt | 0 | 1.1 |
| Fortress | 20.6 | 16 |
| gdpval | 0 | 12.8 |
| GPQA Diamond | 51.5 | 83.3 |
| harmbench | 62.7 | 98.4 |
| HLE | 4.2 | 20.3 |
| IFBench | 39 | 71.4 |
| LiveCodeBench | 30.1 | 84.7 |
| longbench_v2 | 36.1 | 58.8 |
| math_500_em | 73.8 | 98.1 |
| MMLU-Pro | 73.4 | 85 |
| OmniScience Accuracy | 23.2 | 38.6 |
| OmniScience Non-Hallucination | 49 | 12.9 |
| scicode | 29.9 | 41 |
| simple_safety_tests | 98.8 | 99 |
| simplebench | 23 | 53.1 |
| simpleqa | 23.2 | 49.4 |
| SWE-bench Verified | 24.5 | 69.1 |
| Terminal-Bench Hard | 6.8 | 37.1 |
| xstest | 95.9 | 97.3 |
| τ²-Bench Telecom (AA run) | 19 | 80.7 |
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.