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 1634 ± 1, rank #169 of 1392 rated models, from 143 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Stats (IMO-AnswerBench) | 92.3 | Score (%) | 100 |
| LLM Stats (PinchBench) | 90 | Score (%) | 100 |
| LiveBench Simplify | 72.28 | Score | 92.5 |
| OpenRouter Tau2-Bench Airline | 76.9 | Accuracy (%) | 89.5 |
| LiveBench Paraphrase | 77.38 | Score | 86.8 |
| AI for Education Pedagogy - Maths | 88.89 | Accuracy (%) | 85.5 |
| Vals AI LiveCodeBench | 85.98 | Accuracy (%) | 83.8 |
| Onyx Open LLM Leaderboard | 86.8 | MMLU-Pro | 83.3 |
| Conceptual Reasoning Index - Decision Theory (DTBench) | 83.55 | Chance-Corrected Score (0-100) | 83.2 |
| LLM Stats (Multi-Challenge) | 63.8 | Score (%) | 82.1 |
| LLM Stats (LongBench v2) | 61.9 | Score (%) | 81.2 |
| PinchBench | 90.58 | Success Rate (%) | 81 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-ultra · How It Works · Data refreshed daily, snapshot 2026-09-05.