| Benchmark | Claude 3.7 Sonnet | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 37 | 60.2 |
| AA Intelligence | 27.6 | 43.3 |
| AA-LCR | 62.3 | 79.3 |
| AA-Omniscience | -0.7 | 0.3 |
| AIME 2025 | 54.8 | 100 |
| arc_agi_1 | 13.6 | 86.2 |
| ARC-AGI-2 | 0 | 52.9 |
| arena_vision | 1195 | 1245 |
| Artificial Analysis Coding Index | 36.4 | 48.7 |
| critpt | 0.9 | 11.6 |
| Fortress | 38 | 17.5 |
| gdpval | 27.4 | 48.3 |
| GPQA Diamond | 84.8 | 92.4 |
| HLE | 9.7 | 45.5 |
| hmmt_feb_2025 | 31.7 | 100 |
| IFBench | 48.3 | 75.4 |
| metr_hcast | 54.2 | 77 |
| metr_re_bench | 0 | 35 |
| metr_swaa | 98.7 | 99.6 |
| mmmlu | 86.1 | 89.6 |
| MMMU-Pro | 60.1 | 79.5 |
| MultiNRC | 27.8 | 42.2 |
| OmniScience Accuracy | 28.2 | 44.3 |
| OmniScience Non-Hallucination | 60 | 38.4 |
| scicode | 40.3 | 52.1 |
| SEAL VISTA | 48.2 | 46.6 |
| simplebench | 46.4 | 45.8 |
| swe_bench_bash | 52.8 | 72.8 |
| SWE-bench Verified | 70.3 | 80 |
| Terminal-Bench Hard | 21.2 | 62.2 |
| τ²-Bench Telecom (AA run) | 54.7 | 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.