Across 19 shared benchmarks, Granite 4.2 30B scores higher on 18 and Granite 4.2 8B on 1. The widest gap is SWE-bench Pro, where Granite 4.2 30B scores 33.3 against 19.1.
| Benchmark | Granite 4.2 30B | Granite 4.2 8B |
|---|---|---|
| AIME25 no tools | 89.2 | 86.7 |
| Arena-Hard-V2 | 67.9 | 65.2 |
| BFCL (v4) | 61.4 | 52.4 |
| BirdBench | 41.9 | 41.1 |
| gdpval | 1225 | 1189 |
| GPQA Diamond | 66.4 | 64.1 |
| HMMT Feb25 | 89.2 | 78.3 |
| IFBench (prompt) | 74.7 | 79.3 |
| LiveCodeBench v6 | 75.8 | 73.2 |
| MMLU-Pro | 77.6 | 74 |
| MMLU-ProX lite (IBM) | 66.6 | 61.1 |
| RULER 128K | 81.4 | 71.4 |
| RULER 64K | 90 | 81 |
| scicode | 38.8 | 36.1 |
| SWE-bench Multilingual | 41.9 | 30.8 |
| SWE-bench Pro | 33.3 | 19.1 |
| SWE-bench Verified | 57 | 47.7 |
| Terminal-Bench 2.1 | 29.2 | 20.6 |
| τ³-Bench | 68 | 66.3 |
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.