AI provider / NVIDIA
NVIDIA
A provider strongly associated with the hardware and software stack powering the AI era, and increasingly with model lines built for enterprise acceleration.
NVIDIA powers the AI era through accelerated computing — and increasingly participates in the model layer itself. The Nemotron line started as its own model family and grew into open reasoning models tuned for the way teams package, optimize and deploy AI at scale.
From Nemotron-4 340B for synthetic data to the Llama Nemotron reasoning tiers and the hybrid-MoE Nemotron 3 family, the strategy pairs hardware leadership with open, deployment-ready models — extended toward full multimodality with Nano Omni.
Model catalog
Every release in this provider’s model line.
Each row is a distinct model release — its date, its type and the traits it is known for. Open a source to dig into any of them.
| Model | Released | Type | Key traits | Source |
|---|---|---|---|---|
| Nemotron 3.5 Super VL & Bolt | Oct 2026 | Multimodal |
| |
| Nemotron 3.5 Lightning | Aug 2026 | Open release |
| |
| Nemotron 3 Ultra & Nano Omni | Mar 2026 | Multimodal |
| |
| Nemotron 3 | Dec 2025 | Open family |
| |
| Nemotron Nano 2 | Aug 2025 | Reasoning |
| |
| Llama Nemotron (Nano/Super/Ultra) | Mar 2025 | Reasoning |
| |
| Llama-3.1-Nemotron-70B | Oct 2024 | Alignment |
| |
| Nemotron-4 340B | Jun 2024 | Synthetic data |
| |
| Nemotron-3 8B | Nov 2023 | Debut |
|
Timeline
A practical timeline of model releases and company milestones.
This page tracks the moments that matter most: major model releases, version jumps and company moves that changed how this provider shows up in the AI market.
Nemotron 3.5 Super VL spans data center to desktop in one family.
A 120B vision-language MoE with 12B active parameters leads a trio down to a 4GB edge model, delivering up to 5x the agentic throughput of comparable open models.
Nemotron 3.5 Lightning squeezes into a single consumer GPU.
A 30B MoE with roughly 3B active parameters per token, it trades scale for reach — open and light enough to run locally rather than only in a datacenter.
Nemotron 3 Ultra arrives at GTC, with Nano Omni adding full multimodality.
The top tier ships alongside a 30B omni model that sees, hears and reads — built as a perception sub-agent for agentic AI.
Nemotron 3 launches as a hybrid-MoE open family.
Nano, Super and Ultra tiers deliver 4x the throughput of the previous generation for multi-agent systems at scale.
Nemotron Nano 2 debuts hybrid Mamba-transformer reasoning.
A 9B model built for throughput shows the architecture direction the Nemotron 3 generation will scale up.
Llama Nemotron brings open reasoning in Nano, Super and Ultra tiers.
Announced at GTC for agentic platforms, the family fixes the size-tier naming NVIDIA still uses today.
Llama-3.1-Nemotron-70B beats GPT-4o on alignment benchmarks.
NVIDIA’s post-training recipe applied to Meta’s weights briefly tops Arena Hard, validating the refine-don’t-pretrain approach.
Nemotron-4 340B targets synthetic data generation.
Base, Instruct and Reward models with a license built for generating training data — NVIDIA arms everyone else’s model teams.
Nemotron-3 8B opens NVIDIA’s own model line.
The chip company starts shipping foundation models for its NeMo framework, foreshadowing a models-plus-silicon strategy.