| Benchmark | GPT-5.4 | o1 |
|---|---|---|
| AA Agentic Index | 58.2 | 31.1 |
| AA Intelligence | 53.1 | 24 |
| AA-LCR | 77.7 | 63.3 |
| AA-Omniscience | 5.8 | -10.6 |
| arena_vision | 1277 | 1193 |
| Artificial Analysis Coding Index | 71.1 | 39.7 |
| Codeforces | 3168 | 2061 |
| critpt | 23.4 | 0.3 |
| gdpval | 50.1 | 11.5 |
| GPQA Diamond | 93 | 75.7 |
| HLE | 43.7 | 7.7 |
| IFBench | 73.9 | 70.3 |
| livebench | 80.3 | 52.3 |
| OmniScience Accuracy | 50.9 | 34.7 |
| OmniScience Non-Hallucination | 17.4 | 30.7 |
| scicode | 56.6 | 35.8 |
| SEAL VISTA | 50.9 | 45.3 |
| Terminal-Bench Hard | 57.6 | 12.9 |
| τ²-Bench Telecom (AA run) | 87.1 | 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.