Across 14 shared benchmarks, Gemini 1.5 Pro scores higher on 9 and NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 on 5. The widest gap is OpenBookQA, where Gemini 1.5 Pro scores 95.2 against 48.6.
| Benchmark | Gemini 1.5 Pro | NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16 |
|---|---|---|
| GPQA Diamond | 59.1 | 50 |
| GSM8K | 95.2 | 88.1 |
| hellaswag | 93.3 | 90.5 |
| humaneval | 86.6 | 83.8 |
| MATH | 92 | 82 |
| mgsm | 87.5 | 87.7 |
| mmlu | 86.8 | 89.1 |
| MMLU-Pro | 75.8 | 79.1 |
| OpenBookQA | 95.2 | 48.6 |
| RULER 128K | 91.7 | 92.5 |
| RULER 1M | 85 | 76.8 |
| RULER 256K | 91.6 | 86.2 |
| RULER 512K | 86.1 | 84.5 |
| RULER 64K | 93.8 | 95.3 |
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.