| Benchmark | GPT-4o | Opus 5 |
|---|---|---|
| AA Agentic Index | 9.7 | 59.2 |
| AA Intelligence | 11.2 | 63.1 |
| AA-LCR | 55 | 78.7 |
| AA-Omniscience | -10.5 | 37.1 |
| arc_agi_1 | 4.5 | 97.5 |
| ARC-AGI-2 | 0 | 90.4 |
| Artificial Analysis Coding Index | 24.2 | 78 |
| critpt | 0 | 29.1 |
| gdpval | 0 | 67.2 |
| GPQA Diamond | 70.1 | 93.7 |
| HLE | 5.3 | 64.7 |
| MMMU-Pro | 59.9 | 84.7 |
| OmniScience Accuracy | 23.7 | 60.9 |
| OmniScience Non-Hallucination | 62.1 | 40.5 |
| OSWorld-Verified | 5 | 4 |
| scicode | 33.4 | 55.7 |
| simplebench | 17.8 | 80.6 |
| SWE-bench Verified | 38.8 | 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.