| Benchmark | Claude Sonnet 5 | GPT-5.4 |
|---|---|---|
| AA Agentic Index | 49.7 | 58.2 |
| AA Intelligence | 55.3 | 53.1 |
| AA-LCR | 77 | 77.7 |
| AA-Omniscience | 16.4 | 5.8 |
| arena_elo | 1463 | 1477 |
| arena_text_factuality | 1453 | 1476 |
| Artificial Analysis Coding Index | 71.5 | 71.1 |
| browsecomp | 84.7 | 82.7 |
| charxiv_rq | 77 | 82.8 |
| coding_arena_elo | 1540 | 1457 |
| critpt | 16.9 | 23.4 |
| DeepSWE 1.1 | 54 | 52 |
| gdpval | 54.8 | 50.1 |
| GPQA Diamond | 91.1 | 93 |
| HLE | 57.4 | 43.7 |
| Legal Agent Benchmark | 5.8 | 0.4 |
| livebench_agentic_coding | 59.4 | 53.8 |
| livebench_coding | 80.7 | 77.5 |
| livebench_data_analysis | 71.7 | 79.3 |
| livebench_instruction_following | 63.9 | 70.2 |
| livebench_language | 75 | 82.6 |
| livebench_math | 92.9 | 94.2 |
| livebench_reasoning | 88.7 | 88.1 |
| MMMU-Pro | 77.3 | 81.2 |
| officeqa | 73.3 | 68.1 |
| officeqa_pro | 59.4 | 51.1 |
| OmniScience Accuracy | 40 | 50.9 |
| OmniScience Non-Hallucination | 60.6 | 17.4 |
| OSWorld-Verified | 81.2 | 75 |
| scicode | 53.6 | 56.6 |
| simpleqa_verified | 25 | 45.3 |
| SWE-bench Multilingual | 78.3 | 71.7 |
| SWE-bench Pro | 63.2 | 59.1 |
| TauBench V3 - Banking | 37.3 | 39.6 |
| Terminal-Bench 2.0 | 80.4 | 75.1 |
| Terminal-Bench 2.1 | 80.5 | 78.3 |
| toolathlon | 54.3 | 54.6 |
| usamo_2026 | 79.5 | 95.2 |
| τ³-Bench | 28.2 | 72.9 |
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.