Across 11 shared benchmarks, GPT-5 scores higher on 9 and Granite 4.2 8B on 2. The widest gap is τ³-Bench, where Granite 4.2 8B scores 66.3 against 19.6.
| Benchmark | GPT-5 | Granite 4.2 8B |
|---|---|---|
| AIME25 no tools | 94.6 | 86.7 |
| gdpval | 32.5 | 1189 |
| GPQA Diamond | 87.3 | 64.1 |
| LiveCodeBench v6 | 87 | 73.2 |
| MMLU-Pro | 87.5 | 74 |
| scicode | 43 | 36.1 |
| SWE-bench Multilingual | 55.3 | 30.8 |
| SWE-bench Pro | 41.8 | 19.1 |
| SWE-bench Verified | 74.9 | 47.7 |
| Terminal-Bench 2.1 | 35.2 | 20.6 |
| τ³-Bench | 19.6 | 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.