| Benchmark | Claude Opus 4.7 | DeepSeek-V3 |
|---|---|---|
| AA Agentic Index | 64.6 | 1.6 |
| AA Intelligence | 55 | 15.4 |
| AA-LCR | 75.3 | 41.3 |
| AA-Omniscience | 27.3 | -37.6 |
| Artificial Analysis Coding Index | 73.6 | 23 |
| critpt | 12 | 0 |
| gdpval | 58.6 | 0 |
| GPQA Diamond | 94.2 | 68.4 |
| harmbench | 17.5 | 49.7 |
| HLE | 54.7 | 5.2 |
| IFBench | 58.6 | 41 |
| livebench | 76.9 | 72.4 |
| mmmlu | 91.5 | 79.4 |
| OmniScience Accuracy | 48.9 | 25.4 |
| OmniScience Non-Hallucination | 57.7 | 23.3 |
| scicode | 54.5 | 35.8 |
| simplebench | 61.7 | 27.2 |
| SWE-bench Verified | 87.6 | 42 |
| TauBench V3 - Banking | 34.6 | 4.7 |
| Terminal-Bench 2.1 | 83.1 | 16.9 |
| Terminal-Bench Hard | 54.5 | 15.2 |
| vectara_answer_rate | 98 | 97.5 |
| vectara_avg_summary_length | 149.1 | 81.7 |
| vectara_factual_consistency | 88 | 93.9 |
| vectara_hallucination_rate ↓ | 12 | 6.1 |
| τ²-Bench Telecom (AA run) | 88.6 | 47.1 |
| τ³-Bench | 28.9 | 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.