| Benchmark | GPT-5 | o1 |
|---|---|---|
| AA Agentic Index | 49.7 | 31.1 |
| AA Intelligence | 35.3 | 24 |
| AA-LCR | 76.3 | 63.3 |
| AA-Omniscience | -8.7 | -10.6 |
| aider_polyglot | 88 | 61.7 |
| AIME 2025 | 95 | 81.7 |
| AIR-Bench 2024 | 87.7 | 80 |
| arena_vision | 1212 | 1193 |
| Artificial Analysis Coding Index | 38.9 | 39.7 |
| critpt | 5.7 | 0.3 |
| Fortress | 17 | 19.4 |
| gdpval | 32.5 | 11.5 |
| GPQA Diamond | 87.3 | 75.7 |
| GSM8K | -1.8 | 97.1 |
| harmbench | 70 | 96.3 |
| HLE | 28.5 | 7.7 |
| humaneval | 93.4 | 88.1 |
| IFBench | 73.1 | 70.3 |
| metr_hcast | 68.4 | 42.6 |
| metr_re_bench | 40 | 0 |
| metr_swaa | 100 | 100 |
| MMMU | 84.2 | 77.6 |
| OmniScience Accuracy | 40.3 | 34.7 |
| OmniScience Non-Hallucination | 21 | 30.7 |
| scicode | 43 | 35.8 |
| SEAL VISTA | 49.7 | 45.3 |
| simplebench | 56.7 | 41.7 |
| SWE-bench Verified | 74.9 | 48.9 |
| Terminal-Bench Hard | 37.9 | 12.9 |
| τ²-Bench Telecom (AA run) | 86.5 | 62.6 |
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.