| Benchmark | Claude 3.5 Sonnet | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Intelligence | 10 | 8.5 |
| aider_edit_acc | 84.2 | 63.9 |
| aider_polyglot | 45.3 | 5.8 |
| AIME 2024 | 16 | 23.3 |
| AIR-Bench 2024 | 85.9 | 58.6 |
| anthropic_red_team | 99.8 | 96.5 |
| Arena Hard | 87.6 | 69.3 |
| Artificial Analysis Coding Index | 30.2 | 14.5 |
| bbh | 93.1 | 85.9 |
| bbq | 94.9 | 94.5 |
| C-Eval | 76.7 | 72.5 |
| Chinese SimpleQA (C-SimpleQA) | 56.8 | 54.7 |
| CLUEWSC | 85.4 | 84.7 |
| Codeforces | 717 | 25.3 |
| DROP | 87.1 | 84.8 |
| DROP (3-shot F1) | 88.3 | 88.7 |
| DROP (F1) | 88.8 | 92.5 |
| Fortress | 13 | 20.6 |
| FRAMES (Acc.) | 72.5 | 70 |
| GPQA Diamond | 67.2 | 51.5 |
| GSM8K | 96.9 | 96.8 |
| harmbench | 98.1 | 62.7 |
| HLE | 3.9 | 4.2 |
| humaneval | 93.7 | 89 |
| HumanEval-Mul (Pass@1) | 81.7 | 77.2 |
| ifeval | 90.1 | 88.6 |
| IFEval (avg) | 90.1 | 86.4 |
| LiveCodeBench | 32.8 | 30.1 |
| livecodebench_pass1cot | 36.3 | 28.4 |
| longbench_v2 | 41 | 36.1 |
| MATH | 81.3 | 82.7 |
| math_500_em | 78.3 | 73.8 |
| MBPP+ (EvalPlus-augmented) | 75.1 | 73 |
| mgsm | 91.6 | 91.6 |
| mmlu | 88.3 | 88.6 |
| mmlu_redux | 88.9 | 86.2 |
| MMLU-Pro | 78 | 73.4 |
| narrativeqa | 74.6 | 74.9 |
| naturalquestions_closedbook | 50.2 | 45.6 |
| OpenBookQA | 97.2 | 94 |
| scicode | 36.6 | 29.9 |
| simple_safety_tests | 100 | 98.8 |
| simplebench | 41.4 | 23 |
| simpleqa | 28.4 | 23.2 |
| SWE-bench Verified | 50.8 | 24.5 |
| xstest | 95.6 | 95.9 |
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.