Nemotron 3 Ultra: benchmark results
NVIDIA's flagship open Nemotron 3 Ultra MoE model (550B total, 55B active). Provider: NVIDIA. Released 2026-06-04. Access: Open.
Aggregate Rating 1685 ± 7, rank #235 of 1656 rated models, from 470 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Jais 3 Model Cards - Arabic Legal (HELM, IFM-audited) | 77 | Accuracy (%) | 100 |
| Jais 3 Model Cards - HalluScore | 71.2 | Accuracy (%) | 100 |
| LLM Stats (OmniScience) | 78.7 | Score (%) | 100 |
| LLM Stats (RULER) | 94.7 | Score (%) | 100 |
| NVIDIA Nemotron 3.5 Super VL Card - AA-Omniscience hallucination rate | 22.1 | Hallucination rate (%, lower is better) | 100 |
| NVIDIA Nemotron 3.5 Super VL Card - IFBench | 82.9 | Score (%) | 100 |
| PinchBench - 147 Tasks - OpenClaw Version Unspecified | 90.58 | Best Official Run Score (%) | 100 |
| AA IFBench | 81.36 | Accuracy (%) | 99.3 |
| Jais 3 Model Cards - AraDiCE-Culture (IFM-audited) | 88.8 | Accuracy (%) | 96.4 |
| Jais 3 Model Cards - SimpleQA Verified (Arabic) - F-score | 32.9 | F-score (%) | 96.4 |
| PACT - Rule-Scope Discernment | 0.89 | Score (0-1) | 95.2 |
| FinRiskAtlas - Evidence-Grounded Processing - Quantitative Reasoning | 81.9 | Numerical Accuracy (%) | 93.8 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-ultra · How It Works · Data refreshed daily, snapshot 2026-10-20.