Across 20 shared benchmarks, Nemotron 3 Ultra 550B A55B Base and NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 record the same score on every one.
| Benchmark | Nemotron 3 Ultra 550B A55B Base | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 |
|---|---|---|
| AGIEval-En | 78.7 | 78.7 |
| arc_challenge | 97.3 | 97.3 |
| GSM8K | 88.1 | 88.1 |
| hellaswag | 90.5 | 90.5 |
| humaneval | 83.8 | 83.8 |
| MATH | 82 | 82 |
| MBPP-Sanitized | 86 | 86 |
| mgsm | 87.7 | 87.7 |
| mmlu | 89.1 | 89.1 |
| MMLU Global Lite | 90.1 | 90.1 |
| MMLU-Pro | 79.1 | 79.1 |
| OpenBookQA | 48.6 | 48.6 |
| PIQA | 83.8 | 83.8 |
| RACE | 92.2 | 92.2 |
| RULER 128K | 92.5 | 92.5 |
| RULER 1M | 76.8 | 76.8 |
| RULER 256K | 86.2 | 86.2 |
| RULER 512K | 84.5 | 84.5 |
| RULER 64K | 95.3 | 95.3 |
| winogrande | 79.3 | 79.3 |
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.