| Benchmark | Claude 4 Opus | DeepSeek-V4-Pro |
|---|---|---|
| AA Intelligence | 31.7 | 53 |
| AA-LCR | 40 | 70 |
| Artificial Analysis Coding Index | 34 | 59.4 |
| GPQA Diamond | 79.6 | 90.5 |
| HLE | 12.3 | 48.2 |
| IFBench | 53.7 | 76.5 |
| livebench | 74.6 | 73.6 |
| LiveCodeBench | 70.4 | 93.5 |
| MMLU-Pro | 86.6 | 87.5 |
| scicode | 40.9 | 50 |
| simplebench | 58.8 | 50.9 |
| simpleqa | 22.8 | 57.9 |
| SWE-bench Verified | 79.4 | 80.6 |
| Terminal-Bench Hard | 31.1 | 46.2 |
| vectara_answer_rate | 91 | 97.2 |
| vectara_avg_summary_length | 123.2 | 153.8 |
| vectara_factual_consistency | 88 | 91.4 |
| vectara_hallucination_rate ↓ | 12 | 8.6 |
| τ²-Bench Telecom (AA run) | 73.4 | 96.2 |
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.