| Benchmark | DeepSeek-V4-Flash | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 61.3 | 40.5 |
| AA Intelligence | 51.8 | 52 |
| AA-LCR | 74.3 | 79 |
| AA-Omniscience | -14.3 | 22.1 |
| aime_2026 | 95.8 | 96.7 |
| arc_agi_1 | 84 | 34.5 |
| ARC-AGI-2 | 46 | 2.6 |
| arena_elo | 1439 | 1481 |
| arena_text_factuality | 1434 | 1470 |
| Artificial Analysis Coding Index | 69.1 | 69.2 |
| coding_arena_elo | 1431 | 1538 |
| critpt | 16.6 | 10.6 |
| DeepSWE | 54.4 | 49 |
| gdpval | 52.9 | 46.1 |
| GDPval-AA v2 | 1189 | 1421 |
| GPQA Diamond | 90.8 | 92.8 |
| HLE | 38.6 | 40.8 |
| hmmt_feb_2026 | 94.8 | 89.4 |
| livebench_agentic_coding | 46.8 | 43.4 |
| livebench_coding | 75 | 77.9 |
| livebench_data_analysis | 79.3 | 63 |
| livebench_instruction_following | 65.5 | 75.4 |
| livebench_language | 79.2 | 83.9 |
| livebench_math | 86.8 | 86.4 |
| livebench_reasoning | 86.6 | 85.2 |
| OmniScience Accuracy | 40.4 | 50 |
| OmniScience Non-Hallucination | 15.6 | 44.4 |
| scicode | 49.9 | 52.7 |
| simpleqa_verified | 45 | 68.7 |
| SWE-bench Pro | 52.6 | 58.7 |
| TauBench V3 - Banking | 39.4 | 29.9 |
| Terminal-Bench 2.1 | 82.7 | 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.