Nemotron 3 Super: benchmark results
NVIDIA's open 120B-total/12B-active hybrid Mamba-Transformer MoE (March 2026) with 1M context, tuned for high-throughput agentic inference. Provider: NVIDIA. Released 2026-03-11. Access: Open.
Unified ELO 1588 ± 1, rank #298 of 1392 rated models, from 119 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Stats (WMT24++) | 86.67 | Score (%) | 100 |
| AGC-Bench - grapheval_ai_researcher | 1.42 | Dataset z-score | 98.8 |
| AGC-Bench - mops | 1.53 | Dataset z-score | 95.1 |
| AGC-Bench - ttcw | 1.68 | Dataset z-score | 93.9 |
| Bullshit Benchmark | 65.5 | BS Detection Rate (%) | 90.4 |
| AGC-Bench - future_ideas | 1.18 | Dataset z-score | 89 |
| AGC-Bench - conceptual_design | 0.98 | Dataset z-score | 87.8 |
| AGC-Bench - rpgbench | 1.09 | Dataset z-score | 86.4 |
| AGC-Bench - STEM | 0.53 | JRT z-score | 84.1 |
| AGC-Bench - thenextchapter | 0.76 | Dataset z-score | 82.9 |
| AGC-Bench - ss_gen | 0.95 | Dataset z-score | 81.5 |
| AGC-Bench - schnovel | 0.98 | Dataset z-score | 81.1 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-super · How It Works · Data refreshed daily, snapshot 2026-09-05.