Across 6 shared benchmarks, Claude Opus 4.6 scores higher on 6 and Granite 4.2 3B on 0. The widest gap is scicode, where Claude Opus 4.6 scores 52 against 24.1.
| Benchmark | Claude Opus 4.6 | Granite 4.2 3B |
|---|---|---|
| AIME25 no tools | 95.6 | 78.3 |
| GPQA Diamond | 91.3 | 54.8 |
| LiveCodeBench v6 | 88.8 | 69.7 |
| MMLU-Pro | 89.1 | 67.8 |
| scicode | 52 | 24.1 |
| τ³-Bench | 72.4 | 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.