| Benchmark | DeepSeek-V4-Pro | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 63.3 | 1 |
| AA Intelligence | 53 | 7 |
| Artificial Analysis Coding Index | 59.4 | 11.4 |
| gdpval | 49 | 0 |
| GPQA Diamond | 90.5 | 42.6 |
| HLE | 48.2 | 4.2 |
| IFBench | 76.5 | 31 |
| LiveCodeBench | 93.5 | 35.5 |
| scicode | 50 | 22.9 |
| simplebench | 50.9 | 10.7 |
| simpleqa | 57.9 | 9.9 |
| SWE-bench Verified | 80.6 | 8.7 |
| TauBench V3 - Banking | 30.1 | 2.9 |
| Terminal-Bench 2.1 | 72.1 | 5.6 |
| τ³-Bench | 25.8 | 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.