Across 14 shared benchmarks, Granite 4.2 3B scores higher on 1 and Granite 4.2 8B on 13. The widest gap is MMLU-ProX lite (IBM), where Granite 4.2 8B scores 61.1 against 27.8.
| Benchmark | Granite 4.2 3B | Granite 4.2 8B |
|---|---|---|
| AIME25 no tools | 78.3 | 86.7 |
| Arena-Hard-V2 | 35 | 65.2 |
| BFCL (v4) | 52.4 | 52.4 |
| GPQA Diamond | 54.8 | 64.1 |
| HMMT Feb25 | 66.7 | 78.3 |
| IFBench (prompt) | 74.3 | 79.3 |
| LiveCodeBench v6 | 69.7 | 73.2 |
| MMLU-Pro | 67.8 | 74 |
| MMLU-ProX lite (IBM) | 27.8 | 61.1 |
| RULER 128K | 55.3 | 71.4 |
| RULER 64K | 67.5 | 81 |
| scicode | 24.1 | 36.1 |
| τ³-Bench | 51 | 66.3 |
| τ³-bench (AVG) | 45.8 | 58.1 |
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.