Across 6 shared benchmarks, GPT-5 scores higher on 5 and Granite 4.2 3B on 1. The widest gap is τ³-Bench, where Granite 4.2 3B scores 51 against 19.6.
| Benchmark | GPT-5 | Granite 4.2 3B |
|---|---|---|
| AIME25 no tools | 94.6 | 78.3 |
| GPQA Diamond | 87.3 | 54.8 |
| LiveCodeBench v6 | 87 | 69.7 |
| MMLU-Pro | 87.5 | 67.8 |
| scicode | 43 | 24.1 |
| τ³-Bench | 19.6 | 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.