| Benchmark | Claude Sonnet 4 | DeepSeek-V4-Pro |
|---|---|---|
| AA Agentic Index | 39.2 | 63.3 |
| AA Intelligence | 29.8 | 53 |
| AA-LCR | 69.7 | 70 |
| AA-Omniscience | 0.2 | -10.6 |
| Artificial Analysis Coding Index | 37.6 | 59.4 |
| browsecomp | 12.2 | 83.4 |
| critpt | 1.1 | 13 |
| gdpval | 31.2 | 49 |
| GPQA Diamond | 78 | 90.5 |
| HLE | 10.7 | 48.2 |
| HLE (with tools) | 20.3 | 48.2 |
| IFBench | 55 | 76.5 |
| livebench | 74.8 | 73.6 |
| LiveCodeBench | 68.5 | 93.5 |
| MMLU-Pro | 84 | 87.5 |
| OmniScience Accuracy | 22.7 | 43 |
| OmniScience Non-Hallucination | 70.9 | 12.2 |
| scicode | 40 | 50 |
| simplebench | 45.5 | 50.9 |
| simpleqa | 15.9 | 57.9 |
| SWE-bench Multilingual | 56.9 | 76.2 |
| SWE-bench Pro | 42.7 | 55.4 |
| SWE-bench Verified | 80.2 | 80.6 |
| TauBench V3 - Banking | 16.7 | 30.1 |
| Terminal-Bench 2.1 | 36.3 | 72.1 |
| Terminal-Bench Hard | 31.1 | 46.2 |
| vectara_answer_rate | 98.6 | 97.2 |
| vectara_avg_summary_length | 145.8 | 153.8 |
| vectara_factual_consistency | 89.7 | 91.4 |
| vectara_hallucination_rate ↓ | 10.3 | 8.6 |
| τ²-Bench Telecom (AA run) | 64.6 | 96.2 |
| τ³-Bench | 13.8 | 25.8 |
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.