MongoDB - Build AI That Scales
Mini toolsNemotron 3 Super
By NVIDIA
Nemotron 3 Super is released as a 12B-active, 120B-total parameter hybrid MoE model that introduces LatentMoE for accuracy, multi-token prediction layers for faster inference, and NVFP4 pretraining, with support for up to 1M context length and published checkpoints plus training datasets.
Text
Gen 7
Overview
Nemotron 3 Super is NVIDIA’s open Mixture-of-Experts hybrid Mamba-Transformer model designed for high-throughput agentic workloads and long-context reasoning.
About NVIDIA
Industry:
Computer Hardware Manufacturing
Company Size:
42000
Location:
Santa Clara, California, US
Website:
nvidia.com
Benchmark scores
How Nemotron 3 Super ranks on tracked AI benchmarks. Click any benchmark to see its full leaderboard.
Compare across all benchmarks →Tools using Nemotron 3 Super
Last updated: April 6, 2026
