Across 18 shared benchmarks, GLM-4.5-Base scores higher on 4 and NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 on 13, with 1 level. The widest gap is RULER 64K, where NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 scores 95.3 against 16.1.
| Benchmark | GLM-4.5-Base | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 |
|---|---|---|
| AGIEval-En | 70.1 | 78.7 |
| arc_challenge | 96.6 | 97.3 |
| GPQA Diamond | 34.9 | 50 |
| GSM8K | 90.1 | 88.1 |
| hellaswag | 90.2 | 90.5 |
| humaneval | 79.8 | 83.8 |
| MATH | 62.6 | 82 |
| MBPP-Sanitized | 76.7 | 86 |
| mgsm | 81.3 | 87.7 |
| mmlu | 87.7 | 89.1 |
| MMLU Global Lite | 85.8 | 90.1 |
| MMLU-Pro | 65.8 | 79.1 |
| OpenBookQA | 49.6 | 48.6 |
| PIQA | 85.1 | 83.8 |
| RACE | 92.2 | 92.2 |
| RULER 128K | 0 | 92.5 |
| RULER 64K | 16.1 | 95.3 |
| winogrande | 85.2 | 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.