| Benchmark | Claude Sonnet 4 | GPT-5.5 |
|---|---|---|
| AA Agentic Index | 39.2 | 47.4 |
| AA Intelligence | 29.8 | 56.3 |
| AA-LCR | 69.7 | 79 |
| AA-Omniscience | 0.2 | 20.5 |
| arc_agi_1 | 63.7 | 95 |
| ARC-AGI-2 | 13.6 | 85 |
| arena_vision | 1207 | 1283 |
| Artificial Analysis Coding Index | 37.6 | 74.9 |
| browsecomp | 12.2 | 84.4 |
| critpt | 1.1 | 27.1 |
| Fortress | 24.4 | 16.3 |
| frontiermath_tier_4 | 0 | 35.4 |
| gdpval | 31.2 | 49.5 |
| GPQA Diamond | 78 | 93.6 |
| HLE | 10.7 | 52.2 |
| HLE (with tools) | 20.3 | 52.2 |
| IFBench | 55 | 75.9 |
| livebench | 74.8 | 80.7 |
| MMMU-Pro | 62.4 | 83.2 |
| OmniScience Accuracy | 22.7 | 58 |
| OmniScience Non-Hallucination | 70.9 | 12 |
| OSWorld-Verified | 42.2 | 78.7 |
| scicode | 40 | 56.1 |
| simplebench | 45.5 | 69 |
| SWE-bench Pro | 42.7 | 58.6 |
| SWE-bench Verified | 80.2 | 80.6 |
| TauBench V3 - Banking | 16.7 | 39 |
| Terminal-Bench 2.1 | 36.3 | 84.3 |
| Terminal-Bench Hard | 31.1 | 60.6 |
| vectara_answer_rate | 98.6 | 100 |
| vectara_avg_summary_length | 145.8 | 129.6 |
| vectara_factual_consistency | 89.7 | 90.7 |
| vectara_hallucination_rate ↓ | 10.3 | 9.3 |
| τ²-Bench Telecom (AA run) | 64.6 | 93.9 |
| τ³-Bench | 13.8 | 31.3 |
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.