| Benchmark | GPT-5 | Opus 5 |
|---|---|---|
| AA Agentic Index | 49.7 | 59.2 |
| AA Intelligence | 35.3 | 63.1 |
| AA-LCR | 76.3 | 78.7 |
| AA-Omniscience | -8.7 | 37.1 |
| arc_agi_1 | 56.2 | 97.5 |
| ARC-AGI-2 | 7.5 | 90.4 |
| Artificial Analysis Coding Index | 38.9 | 78 |
| browsecomp | 54.9 | 90.8 |
| coding_arena_elo | 1419 | 1662 |
| critpt | 5.7 | 29.1 |
| gdpval | 32.5 | 67.2 |
| GPQA Diamond | 87.3 | 93.7 |
| HealthBench | 67.2 | 67.1 |
| healthbench_length_adjusted | 57.7 | 57.8 |
| HLE | 28.5 | 64.7 |
| HLE (with tools) | 35.2 | 64.7 |
| MMMU-Pro | 78.4 | 84.7 |
| OmniScience Accuracy | 40.3 | 60.9 |
| OmniScience Non-Hallucination | 21 | 40.5 |
| scicode | 43 | 55.7 |
| simplebench | 56.7 | 80.6 |
| simpleqa_verified | 50.6 | 56.7 |
| SWE-bench Multilingual | 55.3 | 89.5 |
| SWE-bench Pro | 41.8 | 79.2 |
| SWE-bench Verified | 74.9 | 96 |
| TauBench V3 - Banking | 22.1 | 44.7 |
| Terminal-Bench 2.1 | 35.2 | 89.1 |
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.