| Benchmark | Claude 4 Opus | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Intelligence | 31.7 | 8.5 |
| AA-LCR | 40 | 24.7 |
| aider_polyglot | 70.7 | 5.8 |
| AIME 2024 | 76 | 23.3 |
| AIR-Bench 2024 | 85.7 | 58.6 |
| anthropic_red_team | 98.9 | 96.5 |
| Artificial Analysis Coding Index | 34 | 14.5 |
| bbq | 99.3 | 94.5 |
| Fortress | 27.6 | 20.6 |
| GPQA Diamond | 79.6 | 51.5 |
| harmbench | 91.7 | 62.7 |
| HLE | 12.3 | 4.2 |
| IFBench | 53.7 | 39 |
| ifeval | 87.4 | 88.6 |
| LiveCodeBench | 70.4 | 30.1 |
| longbench_v2 | 55.6 | 36.1 |
| math_500_em | 98.2 | 73.8 |
| mmlu | 92.9 | 88.6 |
| mmlu_redux | 94.2 | 86.2 |
| MMLU-Pro | 86.6 | 73.4 |
| mmmlu | 88.8 | 73.8 |
| scicode | 40.9 | 29.9 |
| simple_safety_tests | 100 | 98.8 |
| simplebench | 58.8 | 23 |
| simpleqa | 22.8 | 23.2 |
| SWE-bench Verified | 79.4 | 24.5 |
| Terminal-Bench Hard | 31.1 | 6.8 |
| xstest | 97 | 95.9 |
| τ²-Bench Telecom (AA run) | 73.4 | 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.