Across 18 shared benchmarks, DeepSeek-V3.2-Exp-Base scores higher on 5 and NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 on 13. The widest gap is MBPP-Sanitized, where NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 scores 86 against 58.7.
| Benchmark | DeepSeek-V3.2-Exp-Base | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 |
|---|---|---|
| AGIEval-En | 70.1 | 78.7 |
| arc_challenge | 95.5 | 97.3 |
| GPQA Diamond | 52 | 50 |
| GSM8K | 91.1 | 88.1 |
| hellaswag | 89.4 | 90.5 |
| humaneval | 67.7 | 83.8 |
| MATH | 62.5 | 82 |
| MBPP-Sanitized | 58.7 | 86 |
| mgsm | 82.3 | 87.7 |
| mmlu | 87.8 | 89.1 |
| MMLU Global Lite | 85.6 | 90.1 |
| MMLU-Pro | 63.3 | 79.1 |
| OpenBookQA | 48.2 | 48.6 |
| PIQA | 85.1 | 83.8 |
| RACE | 93.2 | 92.2 |
| RULER 128K | 91.9 | 92.5 |
| RULER 64K | 93.3 | 95.3 |
| winogrande | 85.6 | 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.