Across 13 shared benchmarks, Granite 4.2 30B scores higher on 13 and Granite 4.2 3B on 0. The widest gap is MMLU-ProX lite (IBM), where Granite 4.2 30B scores 66.6 against 27.8.
| Benchmark | Granite 4.2 30B | Granite 4.2 3B |
|---|---|---|
| AIME25 no tools | 89.2 | 78.3 |
| Arena-Hard-V2 | 67.9 | 35 |
| BFCL (v4) | 61.4 | 52.4 |
| GPQA Diamond | 66.4 | 54.8 |
| HMMT Feb25 | 89.2 | 66.7 |
| IFBench (prompt) | 74.7 | 74.3 |
| LiveCodeBench v6 | 75.8 | 69.7 |
| MMLU-Pro | 77.6 | 67.8 |
| MMLU-ProX lite (IBM) | 66.6 | 27.8 |
| RULER 128K | 81.4 | 55.3 |
| RULER 64K | 90 | 67.5 |
| scicode | 38.8 | 24.1 |
| τ³-Bench | 68 | 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.