| Benchmark | Llama 3.1 Instruct 405B | o1 |
|---|---|---|
| AA Agentic Index | 6.3 | 31.1 |
| AA Intelligence | 8.5 | 24 |
| AA-LCR | 24.7 | 63.3 |
| AA-Omniscience | -17.1 | -10.6 |
| aider_polyglot | 5.8 | 61.7 |
| AIME 2024 | 23.3 | 79.2 |
| AIR-Bench 2024 | 58.6 | 80 |
| anthropic_red_team | 96.5 | 98.3 |
| Artificial Analysis Coding Index | 14.5 | 39.7 |
| bbq | 94.5 | 97.3 |
| Codeforces | 25.3 | 2061 |
| critpt | 0 | 0.3 |
| DROP (3-shot F1) | 88.7 | 90.2 |
| Fortress | 20.6 | 19.4 |
| gdpval | 0 | 11.5 |
| GPQA Diamond | 51.5 | 75.7 |
| GSM8K | 96.8 | 97.1 |
| harmbench | 62.7 | 96.3 |
| HLE | 4.2 | 7.7 |
| humaneval | 89 | 88.1 |
| IFBench | 39 | 70.3 |
| livecodebench_pass1cot | 28.4 | 63.4 |
| MATH | 82.7 | 96.4 |
| math_500_em | 73.8 | 96.4 |
| mgsm | 91.6 | 90.8 |
| mmlu | 88.6 | 91.8 |
| mmmlu | 73.8 | 87.7 |
| OmniScience Accuracy | 23.2 | 34.7 |
| OmniScience Non-Hallucination | 49 | 30.7 |
| scicode | 29.9 | 35.8 |
| simple_safety_tests | 98.8 | 99 |
| simplebench | 23 | 41.7 |
| simpleqa | 23.2 | 47 |
| SWE-bench Verified | 24.5 | 48.9 |
| Terminal-Bench Hard | 6.8 | 12.9 |
| xstest | 95.9 | 97 |
| τ²-Bench Telecom (AA run) | 19 | 62.6 |
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.