| Benchmark | DeepSeek-V4-Flash | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 61.3 | 1 |
| AA Intelligence | 51.8 | 7 |
| ARC-AGI-2 | 46 | 0 |
| Artificial Analysis Coding Index | 69.1 | 11.4 |
| gdpval | 52.9 | 0 |
| GPQA Diamond | 90.8 | 42.6 |
| HLE | 38.6 | 4.2 |
| IFBench | 79.2 | 31 |
| LiveCodeBench | 91.6 | 35.5 |
| scicode | 49.9 | 22.9 |
| simplebench | 46.3 | 10.7 |
| SWE-bench Verified | 79 | 8.7 |
| TauBench V3 - Banking | 39.4 | 2.9 |
| Terminal-Bench 2.1 | 82.7 | 5.6 |
| τ³-Bench | 22.9 | 2.9 |
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.