| Benchmark | GPT-4o | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 9.7 | 60.2 |
| AA Intelligence | 11.2 | 43.3 |
| AA-LCR | 55 | 79.3 |
| AA-Omniscience | -10.5 | 0.3 |
| AIME 2025 | 11.7 | 100 |
| arc_agi_1 | 4.5 | 86.2 |
| ARC-AGI-2 | 0 | 52.9 |
| arena_vision | 1162 | 1245 |
| Artificial Analysis Coding Index | 24.2 | 48.7 |
| critpt | 0 | 11.6 |
| Fortress | 47.2 | 17.5 |
| gaia_level1 | 39.8 | 92.5 |
| gaia_level2 | 27 | 74.8 |
| gaia_level3 | 14.3 | 57.1 |
| gaia_overall | 28.9 | 77.4 |
| gdpval | 0 | 48.3 |
| GPQA Diamond | 70.1 | 92.4 |
| HLE | 5.3 | 45.5 |
| IFBench | 36 | 75.4 |
| LiveCodeBench | 38.3 | 89 |
| longbench_v2 | 48.1 | 54.5 |
| mathvision | 30.4 | 83 |
| MathVista | 63.8 | 82.8 |
| metr_hcast | 25 | 77 |
| metr_re_bench | 0 | 35 |
| metr_swaa | 99.2 | 99.6 |
| MMLU-Pro | 74.7 | 87 |
| mmmlu | 81.4 | 89.6 |
| MMMU-Pro | 59.9 | 79.5 |
| MultiNRC | 12.4 | 42.2 |
| OCRBench | 806 | 80.7 |
| OmniScience Accuracy | 23.7 | 44.3 |
| OmniScience Non-Hallucination | 62.1 | 38.4 |
| PropensityBench | 46.1 | 34.4 |
| scicode | 33.4 | 52.1 |
| SEAL VISTA | 34.9 | 46.6 |
| simplebench | 17.8 | 45.8 |
| swe_bench_bash | 21.6 | 72.8 |
| SWE-bench Verified | 38.8 | 80 |
| Tau2 retail | 63.4 | 82 |
| Terminal-Bench Hard | 8.3 | 62.2 |
| vectara_answer_rate | 93.8 | 100 |
| vectara_avg_summary_length | 86.6 | 186.3 |
| vectara_factual_consistency | 90.4 | 89.2 |
| vectara_hallucination_rate ↓ | 9.6 | 10.8 |
| Video-MME | 77.2 | 86 |
| VideoMMMU | 61.2 | 85.9 |
| τ²-Bench Telecom (AA run) | 28.9 | 98.7 |
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.