| Benchmark | Claude Sonnet 5 | GPT-5.2 |
|---|---|---|
| AA Agentic Index | 49.7 | 60.2 |
| AA Intelligence | 55.3 | 43.3 |
| AA-LCR | 77 | 79.3 |
| AA-Omniscience | 16.4 | 0.3 |
| arena_elo | 1463 | 1437 |
| arena_text_factuality | 1453 | 1442 |
| Artificial Analysis Coding Index | 71.5 | 48.7 |
| browsecomp | 84.7 | 65.8 |
| charxiv_rq | 77 | 82.1 |
| coding_arena_elo | 1540 | 1418 |
| critpt | 16.9 | 11.6 |
| gdpval | 54.8 | 48.3 |
| GPQA Diamond | 91.1 | 92.4 |
| HLE | 57.4 | 45.5 |
| livebench_agentic_coding | 59.4 | 50.3 |
| livebench_coding | 80.7 | 76.1 |
| livebench_data_analysis | 71.7 | 78.2 |
| livebench_instruction_following | 63.9 | 61.8 |
| livebench_language | 75 | 79.8 |
| livebench_math | 92.9 | 93.2 |
| livebench_reasoning | 88.7 | 83.2 |
| MMMU-Pro | 77.3 | 79.5 |
| mrcr_v2_8needle_128k_average | 81.5 | 83.8 |
| OmniScience Accuracy | 40 | 44.3 |
| OmniScience Non-Hallucination | 60.6 | 38.4 |
| scicode | 53.6 | 52.1 |
| simplebench | 60.6 | 45.8 |
| simpleqa_verified | 25 | 38.9 |
| SWE-bench Multilingual | 78.3 | 72 |
| SWE-bench Pro | 63.2 | 55.6 |
| SWE-bench Verified | 85.2 | 80 |
| TauBench V3 - Banking | 37.3 | 11.1 |
| Terminal-Bench 2.0 | 80.4 | 54 |
| toolathlon | 54.3 | 46.3 |
| τ³-Bench | 28.2 | 12 |
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.