| Benchmark | GPT-4o mini | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 1 | 6.3 |
| AA Intelligence | 7 | 8.5 |
| aider_polyglot | 3.6 | 5.8 |
| AIR-Bench 2024 | 56.3 | 58.6 |
| anthropic_red_team | 98.3 | 96.5 |
| Artificial Analysis Coding Index | 11.4 | 14.5 |
| bbq | 88.2 | 94.5 |
| DROP | 79.7 | 84.8 |
| Fortress | 48.1 | 20.6 |
| gdpval | 0 | 0 |
| GPQA Diamond | 42.6 | 51.5 |
| GSM8K | 84.3 | 96.8 |
| harmbench | 84.9 | 62.7 |
| HLE | 4.2 | 4.2 |
| humaneval | 87.2 | 89 |
| IFBench | 31 | 39 |
| LiveCodeBench | 35.5 | 30.1 |
| MATH | 80.2 | 82.7 |
| mgsm | 87 | 91.6 |
| mmlu | 81.8 | 88.6 |
| narrativeqa | 76.8 | 74.9 |
| naturalquestions_closedbook | 38.5 | 45.6 |
| OpenBookQA | 92 | 94 |
| scicode | 22.9 | 29.9 |
| simple_safety_tests | 97.8 | 98.8 |
| simplebench | 10.7 | 23 |
| simpleqa | 9.9 | 23.2 |
| SWE-bench Verified | 8.7 | 24.5 |
| xstest | 96 | 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.