Last updated:
Nemotron
NVIDIA's open Nemotron models — large mixture-of-experts models built for long-running AI agents and enterprise deployment.
Model Facts
Latest model: Nemotron 3.5 Lightning
- Provider
- NVIDIA
- Released
- Jun 4, 2026
- Weights
- Open
Full Model List
| Model | Released | Status | Note |
|---|---|---|---|
| Nemotron 3.5 Lightning | Aug 1, 2026 | current | 30B-parameter MoE for always-on agents (code review, tool use, security monitoring); date approximate |
| Nemotron 3 Ultra | Jun 4, 2026 | current | 550B-parameter MoE (55B active) for orchestrating long-running agents |
| Nemotron 4 | – | preview | 1T-parameter open model in development |
- Official model page:
- nvidia.com/en-us/ai-data-science/foundation-models/nemotron
- Official documentation:
- developer.nvidia.com/nemotron
- Other sources:
- Nemotron 3 Ultra (Pillitteri)
- Nemotron 3 Ultra (Tech Insider)
What it is
Nemotron is NVIDIA's family of open models. Nemotron 3 Ultra (June 4, 2026) is a 550B-parameter mixture-of-experts model with 55B active parameters, released under the OpenMDW license and built to orchestrate complex, long-running agents. Nemotron 3.5 Lightning (August 2026) is a 30B-parameter MoE for always-on agent tasks. NVIDIA says it is moving to a release every four to six weeks, and Nemotron 4 is in development.
Where this information comes from
Details come from launch coverage. Check NVIDIA's model cards for exact license terms and context windows. Release dates, context windows and benchmark scores are vendor-reported unless stated, and harnesses differ between labs, so treat the numbers as indicative.
Notable traits
- Open weights from the leading AI chip vendor.
- Optimised for agentic workloads.
- Rapid release cadence.