| Benchmark | DeepSeek-V4-Flash | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 61.3 | 70.4 |
| AA Intelligence | 51.8 | 52 |
| AA-LCR | 74.3 | 81 |
| AA-Omniscience | -14.3 | 21.2 |
| aime_2026 | 95.8 | 95 |
| arc_agi_1 | 84 | 92.5 |
| ARC-AGI-2 | 46 | 72.1 |
| arena_elo | 1439 | 1477 |
| arena_text_factuality | 1434 | 1466 |
| Artificial Analysis Coding Index | 69.1 | 70.1 |
| coding_arena_elo | 1431 | 1490 |
| critpt | 16.6 | 13.1 |
| DeepSWE | 54.4 | 37 |
| gdpval | 52.9 | 57.8 |
| GDPval-AA (Elo) | 1395 | 1656 |
| GDPval-AA v2 | 1189 | 1357 |
| GPQA Diamond | 90.8 | 92.2 |
| HLE | 38.6 | 42.7 |
| hmmt_feb_2026 | 94.8 | 95.5 |
| IFBench | 79.2 | 76.3 |
| itbenchSre | 31.5 | 40.3 |
| livebench_agentic_coding | 46.8 | 49 |
| livebench_coding | 75 | 78.2 |
| livebench_data_analysis | 79.3 | 64.9 |
| livebench_instruction_following | 65.5 | 75.6 |
| livebench_language | 79.2 | 84.6 |
| livebench_math | 86.8 | 88.2 |
| livebench_reasoning | 86.6 | 82 |
| MCP Atlas | 69 | 83.6 |
| OmniScience Accuracy | 40.4 | 51.4 |
| OmniScience Non-Hallucination | 15.6 | 38.2 |
| scicode | 49.9 | 53.1 |
| simplebench | 46.3 | 76.7 |
| simpleqa_verified | 45 | 68.4 |
| SWE-bench Pro | 52.6 | 55.1 |
| TauBench V3 - Banking | 39.4 | 32.2 |
| Terminal-Bench 2.0 | 56.9 | 76.2 |
| Terminal-Bench 2.1 | 82.7 | 78.7 |
| Terminal-Bench Hard | 38.6 | 46.2 |
| toolathlon | 70.3 | 56.5 |
| τ²-Bench Telecom (AA run) | 95.6 | 95.6 |
| τ³-Bench | 22.9 | 25.4 |
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.