| Benchmark | Claude 4 Opus | GPT-4o mini |
|---|---|---|
| AA Intelligence | 31.7 | 7 |
| aider_polyglot | 70.7 | 3.6 |
| AIR-Bench 2024 | 85.7 | 56.3 |
| anthropic_red_team | 98.9 | 98.3 |
| ARC-AGI-2 | 8.6 | 0 |
| Artificial Analysis Coding Index | 34 | 11.4 |
| bbq | 99.3 | 88.2 |
| Fortress | 27.6 | 48.1 |
| GPQA Diamond | 79.6 | 42.6 |
| harmbench | 91.7 | 84.9 |
| HLE | 12.3 | 4.2 |
| IFBench | 53.7 | 31 |
| LiveCodeBench | 70.4 | 35.5 |
| livecodebench_easy | 99.1 | 81.7 |
| livecodebench_hard | 33.1 | 4.3 |
| livecodebench_medium | 80.4 | 25.2 |
| mmlu | 92.9 | 81.8 |
| MMMU | 73.7 | 59.4 |
| scicode | 40.9 | 22.9 |
| simple_safety_tests | 100 | 97.8 |
| simplebench | 58.8 | 10.7 |
| simpleqa | 22.8 | 9.9 |
| SWE-bench Verified | 79.4 | 8.7 |
| xstest | 97 | 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.