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.
Unified ELO 1712 ± 34, rank #253 of 1776 rated models, from 84 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Agent Arena - Steerability | 21.37 | Steerability (%) | 100 |
| LLM Stats (IMO-AnswerBench) | 92.3 | Score (%) | 100 |
| Agent Arena - Confirmed Success | 15.08 | Confirmed Success (%) | 97.3 |
| Agent Arena - Tool Hallucination | 0.09 | Tool Hallucination (%) | 97.3 |
| IOI | 570 | Score (self-reported) | 96.2 |
| Agent Arena | 13.5 | Net Improvement (%) | 94.6 |
| Agent Arena - Bash Recovery | 18.77 | Bash Recovery (%) | 94.6 |
| Vals AI LiveCodeBench | 85.98 | Accuracy (%) | 89 |
| LLM Stats (MMLU-ProX) | 83 | Score (%) | 87.1 |
| LLM Stats (LongBench v2) | 61.9 | Score (%) | 86.7 |
| CritPt | 3.1 | Accuracy (self-reported) | 86.4 |
| AA-LCR | 65.4 | Score (self-reported) | 85.5 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-ultra · How the rankings work · Data refreshed daily, snapshot 2026-07-22.