| Benchmark | DeepSeek-V4-Flash | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 61.3 | 49.7 |
| AA Intelligence | 51.8 | 38.7 |
| AA-LCR | 74.3 | 73 |
| AA-Omniscience | -14.3 | 10.1 |
| aime_2026 | 95.8 | 96.7 |
| arc_agi_1 | 84 | 84.7 |
| ARC-AGI-2 | 46 | 33.6 |
| arena_elo | 1439 | 1473 |
| arena_text_factuality | 1434 | 1452 |
| Artificial Analysis Coding Index | 69.1 | 42.6 |
| coding_arena_elo | 1431 | 1387 |
| critpt | 16.6 | 8.6 |
| gdpval | 52.9 | 35.2 |
| GDPval-AA (Elo) | 1395 | 1204 |
| GPQA Diamond | 90.8 | 90.4 |
| HLE | 38.6 | 43.5 |
| hmmt_feb_2026 | 94.8 | 89.4 |
| IFBench | 79.2 | 78 |
| MCP Atlas | 69 | 62 |
| OmniScience Accuracy | 40.4 | 53.4 |
| OmniScience Non-Hallucination | 15.6 | 7.6 |
| scicode | 49.9 | 50.6 |
| simplebench | 46.3 | 61.1 |
| simpleqa_verified | 45 | 67.4 |
| SWE-bench Multilingual | 73.3 | 72.7 |
| SWE-bench Pro | 52.6 | 49.6 |
| SWE-bench Verified | 79 | 78 |
| TauBench V3 - Banking | 39.4 | 20.8 |
| Terminal-Bench 2.0 | 56.9 | 47.6 |
| Terminal-Bench 2.1 | 82.7 | 58 |
| Terminal-Bench Hard | 38.6 | 38.6 |
| toolathlon | 70.3 | 49.4 |
| τ²-Bench Telecom (AA run) | 95.6 | 80.4 |
| τ³-Bench | 22.9 | 17.5 |
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.