| Benchmark | Claude Sonnet 4.6 | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 61.6 | 1.6 |
| AA Intelligence | 48.4 | 15.4 |
| AA-LCR | 74 | 41.3 |
| AA-Omniscience | 12.2 | -37.6 |
| Artificial Analysis Coding Index | 63 | 23 |
| critpt | 3.1 | 0 |
| gdpval | 54.8 | 0 |
| GPQA Diamond | 89.9 | 68.4 |
| harmbench | 24.2 | 49.7 |
| HLE | 49 | 5.2 |
| IFBench | 56.6 | 41 |
| livebench | 75.5 | 72.4 |
| mmmlu | 89.3 | 79.4 |
| OmniScience Accuracy | 40.9 | 25.4 |
| OmniScience Non-Hallucination | 51.6 | 23.3 |
| scicode | 47 | 35.8 |
| SWE-bench Verified | 79.6 | 42 |
| Tau2 retail | 91.7 | 69.1 |
| TauBench V3 - Banking | 34.4 | 4.7 |
| Terminal-Bench 2.1 | 71.2 | 16.9 |
| Terminal-Bench Hard | 59.1 | 15.2 |
| vectara_answer_rate | 99.9 | 97.5 |
| vectara_avg_summary_length | 114.7 | 81.7 |
| vectara_factual_consistency | 89.4 | 93.9 |
| vectara_hallucination_rate ↓ | 10.6 | 6.1 |
| τ²-Bench Telecom (AA run) | 97.9 | 47.1 |
| τ³-Bench | 30.5 | 4.7 |
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.