| Benchmark | DeepSeek-V4-Pro | GPT-4o |
|---|---|---|
| AA Agentic Index | 63.3 | 9.7 |
| AA Intelligence | 53 | 11.2 |
| AA-LCR | 70 | 55 |
| AA-Omniscience | -10.6 | -10.5 |
| Artificial Analysis Coding Index | 59.4 | 24.2 |
| Chinese SimpleQA (C-SimpleQA) | 77.7 | 64.6 |
| Codeforces | 3206 | 759 |
| critpt | 13 | 0 |
| gdpval | 49 | 0 |
| GPQA Diamond | 90.5 | 70.1 |
| HLE | 48.2 | 5.3 |
| IFBench | 76.5 | 36 |
| LiveCodeBench | 93.5 | 38.3 |
| MMLU-Pro | 87.5 | 74.7 |
| OmniScience Accuracy | 43 | 23.7 |
| OmniScience Non-Hallucination | 12.2 | 62.1 |
| scicode | 50 | 33.4 |
| simplebench | 50.9 | 17.8 |
| simpleqa | 57.9 | 39.4 |
| SWE-bench Verified | 80.6 | 38.8 |
| Terminal-Bench Hard | 46.2 | 8.3 |
| vectara_answer_rate | 97.2 | 93.8 |
| vectara_avg_summary_length | 153.8 | 86.6 |
| vectara_factual_consistency | 91.4 | 90.4 |
| vectara_hallucination_rate ↓ | 8.6 | 9.6 |
| τ²-Bench Telecom (AA run) | 96.2 | 28.9 |
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.