| Benchmark | DeepSeek-V3.2 | GPT-4o mini |
|---|---|---|
| AA Agentic Index | 39.8 | 1 |
| AA Intelligence | 32.8 | 7 |
| ARC-AGI-2 | 4 | 0 |
| Artificial Analysis Coding Index | 44.2 | 11.4 |
| gaia_level1 | 90.3 | 7.5 |
| gaia_level2 | 73.6 | 4.4 |
| gaia_level3 | 61.2 | 0 |
| gaia_overall | 76.7 | 4.7 |
| gdpval | 18.8 | 0 |
| GPQA Diamond | 84 | 42.6 |
| HLE | 40.8 | 4.2 |
| IFBench | 61 | 31 |
| LiveCodeBench | 86 | 35.5 |
| scicode | 39 | 22.9 |
| SWE-bench Verified | 73.1 | 8.7 |
| TauBench V3 - Banking | 18.8 | 2.9 |
| Terminal-Bench 2.1 | 46.8 | 5.6 |
| τ³-Bench | 69.2 | 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.