| Benchmark | Claude Opus 4.6 | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 67.6 | 1.6 |
| AA Intelligence | 44.9 | 15.4 |
| AA-LCR | 74.3 | 41.3 |
| AA-Omniscience | 13.7 | -37.6 |
| AIME 2025 | 99.8 | 51.3 |
| Artificial Analysis Coding Index | 48.1 | 23 |
| Chinese SimpleQA (C-SimpleQA) | 76.4 | 68 |
| critpt | 12.6 | 0 |
| gdpval | 55.9 | 0 |
| GPQA Diamond | 91.3 | 68.4 |
| GSM8K | 0.6 | 96.7 |
| HLE | 53.1 | 5.2 |
| IFBench | 62.5 | 41 |
| livebench | 76.3 | 72.4 |
| LiveCodeBench | 88.8 | 49.6 |
| LiveCodeBench v6 | 88.8 | 46.9 |
| MMLU-Pro | 89.1 | 81.2 |
| mmmlu | 91.1 | 79.4 |
| multichallenge | 56 | 31.4 |
| OmniScience Accuracy | 47 | 25.4 |
| OmniScience Non-Hallucination | 37.2 | 23.3 |
| scicode | 52 | 35.8 |
| simplebench | 67.6 | 27.2 |
| SWE-bench Verified | 80.8 | 42 |
| Tau2 retail | 91.9 | 69.1 |
| Terminal-Bench Hard | 65.4 | 15.2 |
| vectara_answer_rate | 99.8 | 97.5 |
| vectara_avg_summary_length | 137.6 | 81.7 |
| vectara_factual_consistency | 87.8 | 93.9 |
| vectara_hallucination_rate ↓ | 12.2 | 6.1 |
| τ²-Bench Telecom (AA run) | 99.3 | 47.1 |
| τ³-Bench | 72.4 | 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.