Across 15 shared benchmarks, Nemotron 3.5 Lightning scores higher on 1 and NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16 on 0, with 14 level. The widest gap is MMLU-Pro, where Nemotron 3.5 Lightning scores 81.9 against 67.9.
| Benchmark | Nemotron 3.5 Lightning | NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16 |
|---|---|---|
| AGIEval-En | 70 | 70 |
| arc_challenge | 92.7 | 92.7 |
| GSM8K | 91.3 | 91.3 |
| hellaswag | 85.5 | 85.5 |
| humaneval | 77.4 | 77.4 |
| MBPP | 78.6 | 78.6 |
| Minerva Math | 82.8 | 82.8 |
| mmlu | 78.6 | 78.6 |
| MMLU-Pro | 81.9 | 67.9 |
| OpenBookQA | 47.6 | 47.6 |
| PIQA (acc-norm) | 85.2 | 85.2 |
| PIQA (acc) | 83.3 | 83.3 |
| RULER 1M | 69.6 | 69.6 |
| RULER 256K | 76.9 | 76.9 |
| winogrande | 80 | 80 |
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.