NVIDIA Nemotron 3 Nano 4B — benchmark results
NVIDIA's 4B hybrid Mamba-Transformer edge model (March 2026), compressed from Nemotron Nano 9B v2 for local inference on RTX and Jetson. Provider: NVIDIA. Released 2026-03-16. Access: Open.
Unified ELO 1426 ± 45, rank #1100 of 1776 rated models, from 36 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA IFBench | 58.23 | Accuracy (%) | 70.1 |
| AA Terminal-Bench Hard | 6.82 | Accuracy (%) | 37.1 |
| Tau3 Banking | 7.01 | Success Rate (%) | 36.5 |
| AA-LCR | 16.7 | Score (self-reported) | 35.8 |
| AA Omniscience - Software Engineering (SWE) - Java | 14 | Accuracy (%) | 35.2 |
| AA TAU-2 Bench | 28.07 | Accuracy (%) | 34.4 |
| CritPt | 0 | Accuracy (self-reported) | 30.3 |
| Artificial Analysis Intelligence Index | 8.73 | Intelligence Index | 29 |
| AA Humanity's Last Exam | 4.77 | Accuracy (%) | 28.3 |
| AA Long Context Reasoning | 16.67 | Accuracy (%) | 27.6 |
| AA GPQA Diamond | 51.31 | Accuracy (%) | 27.5 |
| AA CritPt | 0 | Accuracy (%) | 27.1 |
Interactive version: theaggregate.ai/model?slug=nvidia-nemotron-3-nano-4b · How the rankings work · Data refreshed daily, snapshot 2026-07-22.