| Benchmark | Claude Opus 4.8 | DeepSeek-R1 |
|---|---|---|
| AA Agentic Index | 49.4 | 3.1 |
| AA Intelligence | 57.3 | 18.6 |
| AA-LCR | 73 | 56 |
| AA-Omniscience | 28.8 | -31.3 |
| arc_agi_1 | 92.5 | 21.2 |
| ARC-AGI-2 | 72.1 | 1.3 |
| Artificial Analysis Coding Index | 74.3 | 24.6 |
| browsecomp | 88.5 | 8.9 |
| critpt | 20.9 | 0.6 |
| Fortress | 18.2 | 74.4 |
| gdpval | 54.2 | 1.5 |
| GPQA Diamond | 93.6 | 81 |
| HLE | 57.9 | 17.7 |
| IFBench | 62.2 | 39 |
| longbench_v2 | 69.1 | 58.3 |
| OmniScience Accuracy | 48.8 | 30.7 |
| OmniScience Non-Hallucination | 60.7 | 10.5 |
| scicode | 53.5 | 35.7 |
| simplebench | 64.8 | 40.8 |
| simpleqa_verified | 39.5 | 27.4 |
| SWE-bench Multilingual | 84.4 | 30.5 |
| SWE-bench Verified | 88.6 | 57.6 |
| TauBench V3 - Banking | 34.2 | 6.4 |
| Terminal-Bench 2.1 | 85 | 19.1 |
| Terminal-Bench Hard | 58.3 | 6.1 |
| τ²-Bench Telecom (AA run) | 94.4 | 11.4 |
| τ³-Bench | 27.6 | 6.4 |
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.