| Benchmark | GPT-5.4 | o3 |
|---|---|---|
| AA Agentic Index | 58.2 | 36.1 |
| AA Intelligence | 53.1 | 31.1 |
| AA-LCR | 77.7 | 73.3 |
| AA-Omniscience | 5.8 | -15.3 |
| arc_agi_1 | 93.7 | 60.8 |
| ARC-AGI-2 | 74 | 6.5 |
| arena_vision | 1277 | 1217 |
| Artificial Analysis Coding Index | 71.1 | 38.4 |
| browsecomp | 82.7 | 49.7 |
| critpt | 23.4 | 1.1 |
| frontiermath_tier_4 | 27.1 | 2.1 |
| gdpval | 50.1 | 12.8 |
| GPQA Diamond | 93 | 83.3 |
| HLE | 43.7 | 20.3 |
| IFBench | 73.9 | 71.4 |
| MMLU-Pro | 87.5 | 85 |
| MMMU-Pro | 81.2 | 76.4 |
| MultiNRC | 58.3 | 45.5 |
| OmniScience Accuracy | 50.9 | 38.6 |
| OmniScience Non-Hallucination | 17.4 | 12.9 |
| scicode | 56.6 | 41 |
| simpleqa_verified | 45.3 | 53 |
| Terminal-Bench Hard | 57.6 | 37.1 |
| VisualToolBench | 29.2 | 13.7 |
| τ²-Bench Telecom (AA run) | 87.1 | 80.7 |
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.