Nemotron 3 Nano 4B FP8: benchmark results

NVIDIA Nemotron 3 Nano 4B is a separate dense hybrid model with 42 blocks, including 21 Mamba blocks, 17 feed-forward blocks and four attention blocks. The publisher reports 3.97B parameters and a 262,144-token context; it is distinct from the Nano 30B-A3B MoE. Provider: NVIDIA. Released 2026-03-16. Access: Open.

Aggregate Rating 1338 ± 31, rank #8011 of 10916 rated models, from 264 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval English NLU - ScaLA EN58.52Matthews correlation coefficient (%)88.6
EuroEval Portuguese NLU - ScaLA PT32.65Matthews correlation coefficient (%)86.6
EuroEval Spanish NLU - CoNLL ES73.24Micro F1 excluding MISC (%)84.1
EuroEval Italian NLU - ScaLA IT35Matthews correlation coefficient (%)82.2
EuroEval Bulgarian Knowledge30.94Knowledge Average Score (%)69.1
EuroEval Spanish NLU - ScaLA ES27.34Matthews correlation coefficient (%)68.9
EuroEval English NLU - CoNLL EN75.01Micro F1 excluding MISC (%)66.6
EuroEval French NLU - Eltec60.38Micro F1 excluding MISC (%)66.3
EuroEval Portuguese Knowledge54.7Knowledge Average Score (%)65.3
EuroEval German NLU - GermEval61.85Micro F1 excluding MISC (%)65.2
EuroEval Italian Knowledge54.22Knowledge Average Score (%)63.7
EuroEval Dutch Knowledge56.65Knowledge Average Score (%)61.8

Interactive version: theaggregate.ai/model?slug=nemotron-3-nano-4b-fp8 · How It Works · Data refreshed daily, snapshot 2026-11-08.