| Benchmark | DeepSeek-V4-Pro | Gemini 2.5 Pro |
|---|---|---|
| AA Agentic Index | 63.3 | 7.2 |
| AA Intelligence | 53 | 27 |
| AA-LCR | 70 | 66 |
| AA-Omniscience | -10.6 | -14.3 |
| arena_elo | 1457 | 1446 |
| arena_text_factuality | 1450 | 1462 |
| Artificial Analysis Coding Index | 59.4 | 46.7 |
| browsecomp | 83.4 | 9.9 |
| coding_arena_elo | 1582 | 1224 |
| critpt | 13 | 2.6 |
| gdpval | 49 | 8.5 |
| Global-MMLU-Lite | 89.3 | 88.6 |
| GPQA Diamond | 90.5 | 84.4 |
| HLE | 48.2 | 22.5 |
| HLE (with tools) | 48.2 | 28.4 |
| hmmt_nov_2025 | 94.4 | 80 |
| IFBench | 76.5 | 49 |
| LiveCodeBench | 93.5 | 82.7 |
| MMLU-Pro | 87.5 | 86 |
| OmniScience Accuracy | 43 | 39 |
| OmniScience Non-Hallucination | 12.2 | 12.6 |
| scicode | 50 | 43 |
| simplebench | 50.9 | 62.4 |
| simpleqa | 57.9 | 50.8 |
| simpleqa_verified | 57.9 | 56 |
| SWE-bench Verified | 80.6 | 67.2 |
| TauBench V3 - Banking | 30.1 | 9.7 |
| Terminal-Bench 2.1 | 72.1 | 28.5 |
| Terminal-Bench Hard | 46.2 | 26.5 |
| vectara_answer_rate | 97.2 | 99.1 |
| vectara_avg_summary_length | 153.8 | 106.4 |
| vectara_factual_consistency | 91.4 | 93 |
| vectara_hallucination_rate ↓ | 8.6 | 7 |
| τ²-Bench Telecom (AA run) | 96.2 | 54.1 |
| τ³-Bench | 25.8 | 9.3 |
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.