| Benchmark | DeepSeek-V4-Pro | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 63.3 | 40.5 |
| AA Intelligence | 53 | 52 |
| AA-LCR | 70 | 79 |
| AA-Omniscience | -10.6 | 22.1 |
| aime_2026 | 96.7 | 96.7 |
| arena_elo | 1457 | 1481 |
| arena_text_factuality | 1450 | 1470 |
| Artificial Analysis Coding Index | 59.4 | 69.2 |
| coding_arena_elo | 1582 | 1538 |
| critpt | 13 | 10.6 |
| DeepSWE | 12.8 | 49 |
| gdpval | 49 | 46.1 |
| GDPval-AA v2 | 1307 | 1421 |
| GPQA Diamond | 90.5 | 92.8 |
| HLE | 48.2 | 40.8 |
| hmmt_feb_2026 | 95.2 | 89.4 |
| livebench_agentic_coding | 42.6 | 43.4 |
| livebench_coding | 70 | 77.9 |
| livebench_data_analysis | 74.5 | 63 |
| livebench_instruction_following | 62.4 | 75.4 |
| livebench_language | 78.1 | 83.9 |
| livebench_math | 90.7 | 86.4 |
| livebench_reasoning | 82.7 | 85.2 |
| OmniScience Accuracy | 43 | 50 |
| OmniScience Non-Hallucination | 12.2 | 44.4 |
| scicode | 50 | 52.7 |
| simpleqa_verified | 57.9 | 68.7 |
| SWE-bench Pro | 55.4 | 58.7 |
| TauBench V3 - Banking | 30.1 | 29.9 |
| Terminal-Bench 2.1 | 72.1 | 78 |
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.