| Benchmark | Claude 3.5 Sonnet | GPT-4o mini |
|---|---|---|
| AA Intelligence | 10 | 7 |
| AI2D | 94.7 | 75.2 |
| aider_polyglot | 45.3 | 3.6 |
| AIR-Bench 2024 | 85.9 | 56.3 |
| anthropic_red_team | 99.8 | 98.3 |
| Artificial Analysis Coding Index | 30.2 | 11.4 |
| bbq | 94.9 | 88.2 |
| BLINK | 56.5 | 51.9 |
| DROP | 87.1 | 79.7 |
| Fortress | 13 | 48.1 |
| GPQA Diamond | 67.2 | 42.6 |
| GSM8K | 96.9 | 84.3 |
| harmbench | 98.1 | 84.9 |
| HLE | 3.9 | 4.2 |
| humaneval | 93.7 | 87.2 |
| InterGPS (test) | 45.6 | 39.9 |
| LiveCodeBench | 32.8 | 35.5 |
| MATH | 81.3 | 80.2 |
| MathVista | 67.7 | 56.7 |
| mgsm | 91.6 | 87 |
| MMBench (dev-en) | 82.3 | 83.8 |
| mmlu | 88.3 | 81.8 |
| MMMU | 72 | 59.4 |
| MMMU (val) (Pass@1) | 68.3 | 52.1 |
| MMMU-Pro | 54.7 | 41.5 |
| narrativeqa | 74.6 | 76.8 |
| naturalquestions_closedbook | 50.2 | 38.5 |
| OpenBookQA | 97.2 | 92 |
| POPE (test) | 76.6 | 83.6 |
| scicode | 36.6 | 22.9 |
| ScienceQA (img-test) | 73.8 | 84 |
| simple_safety_tests | 100 | 97.8 |
| simplebench | 41.4 | 10.7 |
| simpleqa | 28.4 | 9.9 |
| SWE-bench Verified | 50.8 | 8.7 |
| TextVQA (val) | 70.5 | 70.9 |
| Video-MME Overall | 55.9 | 61.2 |
| xstest | 95.6 | 96 |
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.