| Benchmark | Claude Opus 4.8 | Claude Sonnet 4 |
|---|---|---|
| AA Agentic Index | 49.4 | 39.2 |
| AA Intelligence | 57.3 | 29.8 |
| AA-LCR | 73 | 69.7 |
| AA-Omniscience | 28.8 | 0.2 |
| arc_agi_1 | 92.5 | 63.7 |
| ARC-AGI-2 | 72.1 | 13.6 |
| arena_vision | 1280 | 1207 |
| Artificial Analysis Coding Index | 74.3 | 37.6 |
| browsecomp | 88.5 | 12.2 |
| critpt | 20.9 | 1.1 |
| Fortress | 18.2 | 24.4 |
| frontiermath_tier_4 | 31.3 | 0 |
| gdpval | 54.2 | 31.2 |
| GPQA Diamond | 93.6 | 78 |
| HLE | 57.9 | 10.7 |
| HLE (with tools) | 57.9 | 20.3 |
| IFBench | 62.2 | 55 |
| livebench | 77.2 | 74.8 |
| MMMU-Pro | 78.9 | 62.4 |
| OmniScience Accuracy | 48.8 | 22.7 |
| OmniScience Non-Hallucination | 60.7 | 70.9 |
| OSWorld-Verified | 83.4 | 42.2 |
| scicode | 53.5 | 40 |
| simplebench | 64.8 | 45.5 |
| SWE-bench Multilingual | 84.4 | 56.9 |
| SWE-bench Pro | 69.2 | 42.7 |
| SWE-bench Verified | 88.6 | 80.2 |
| TauBench V3 - Banking | 34.2 | 16.7 |
| Terminal-Bench 2.1 | 85 | 36.3 |
| Terminal-Bench Hard | 58.3 | 31.1 |
| τ²-Bench Telecom (AA run) | 94.4 | 64.6 |
| τ³-Bench | 27.6 | 13.8 |
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.