Across 6 shared benchmarks, DeepSeek-V3.2 scores higher on 6 and Granite 4.2 3B on 0. The widest gap is scicode, where DeepSeek-V3.2 scores 39 against 24.1.
| Benchmark | DeepSeek-V3.2 | Granite 4.2 3B |
|---|---|---|
| AIME25 no tools | 89.3 | 78.3 |
| GPQA Diamond | 84 | 54.8 |
| LiveCodeBench v6 | 83.3 | 69.7 |
| MMLU-Pro | 86 | 67.8 |
| scicode | 39 | 24.1 |
| τ³-Bench | 69.2 | 51 |
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.