Across 11 shared benchmarks, Nemotron 3 Ultra 550B A55B Base scores higher on 9 and NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16 on 2. The widest gap is MMLU-Pro, where Nemotron 3 Ultra 550B A55B Base scores 79.1 against 67.9.
| Benchmark | Nemotron 3 Ultra 550B A55B Base | NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16 |
|---|---|---|
| AGIEval-En | 78.7 | 70 |
| arc_challenge | 97.3 | 92.7 |
| GSM8K | 88.1 | 91.3 |
| hellaswag | 90.5 | 85.5 |
| humaneval | 83.8 | 77.4 |
| mmlu | 89.1 | 78.6 |
| MMLU-Pro | 79.1 | 67.9 |
| OpenBookQA | 48.6 | 47.6 |
| RULER 1M | 76.8 | 69.6 |
| RULER 256K | 86.2 | 76.9 |
| winogrande | 79.3 | 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.