| Benchmark | GPT-4o | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 9.7 | 58.2 |
| AA Intelligence | 11.2 | 53.1 |
| AA-LCR | 55 | 77.7 |
| AA-Omniscience | -10.5 | 5.8 |
| arc_agi_1 | 4.5 | 93.7 |
| ARC-AGI-2 | 0 | 74 |
| arena_vision | 1162 | 1277 |
| Artificial Analysis Coding Index | 24.2 | 71.1 |
| Chinese SimpleQA (C-SimpleQA) | 64.6 | 76.8 |
| Codeforces | 759 | 3168 |
| critpt | 0 | 23.4 |
| gdpval | 0 | 50.1 |
| GPQA Diamond | 70.1 | 93 |
| HLE | 5.3 | 43.7 |
| IFBench | 36 | 73.9 |
| mathvision | 30.4 | 92 |
| MMLU-Pro | 74.7 | 87.5 |
| MMMU-Pro | 59.9 | 81.2 |
| MultiNRC | 12.4 | 58.3 |
| OmniScience Accuracy | 23.7 | 50.9 |
| OmniScience Non-Hallucination | 62.1 | 17.4 |
| OSWorld-Verified | 5 | 75 |
| scicode | 33.4 | 56.6 |
| SEAL VISTA | 34.9 | 50.9 |
| Terminal-Bench Hard | 8.3 | 57.6 |
| vectara_answer_rate | 93.8 | 99.9 |
| vectara_avg_summary_length | 86.6 | 81.7 |
| vectara_factual_consistency | 90.4 | 93 |
| vectara_hallucination_rate ↓ | 9.6 | 7 |
| τ²-Bench Telecom (AA run) | 28.9 | 87.1 |
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.