| Benchmark | DeepSeek-V4-Flash | Gemini 2.5 Flash |
|---|---|---|
| AA Agentic Index | 61.3 | 18.8 |
| AA Intelligence | 51.8 | 17.7 |
| AA-LCR | 74.3 | 65.7 |
| AA-Omniscience | -14.3 | -29.8 |
| arc_agi_1 | 84 | 33.3 |
| ARC-AGI-2 | 46 | 2.5 |
| arena_text_factuality | 1434 | 1437 |
| Artificial Analysis Coding Index | 69.1 | 22.2 |
| critpt | 16.6 | 1.4 |
| gdpval | 52.9 | 11.9 |
| Global-MMLU-Lite | 88.4 | 88.4 |
| GPQA Diamond | 90.8 | 82.8 |
| HLE | 38.6 | 12.1 |
| IFBench | 79.2 | 50.3 |
| LiveCodeBench | 91.6 | 76.2 |
| OmniScience Accuracy | 40.4 | 26.1 |
| OmniScience Non-Hallucination | 15.6 | 24.7 |
| scicode | 49.9 | 35.9 |
| simplebench | 46.3 | 41.2 |
| SWE-bench Verified | 79 | 60.4 |
| Terminal-Bench Hard | 38.6 | 13.6 |
| τ²-Bench Telecom (AA run) | 95.6 | 31.6 |
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.