Ubuntu adds NVIDIA Nemotron 3.5 Lightning in one command
Ubuntu users can now run NVIDIA’s newest open agent model with a single command, removing weeks of setup for teams building always-on AI assistants.
Canonical said the open NVIDIA Nemotron 3.5 Lightning model is available on Ubuntu at launch through pre-packaged inference snaps, deployable across supported NVIDIA hardware with sudo snap install nemotron-3-5-lightning. The model is built for always-on AI agents rather than one-off prompts, a shift that matters to developers who need assistants that persist across long tasks.
The release centers on a 30B hybrid Mixture-of-Experts design with just 3B active parameters, which NVIDIA distilled from its frontier Nemotron 3 Ultra, according to the Ubuntu announcement. Keeping active compute low while total capacity stays high is the trade that lets the model run on workstations, edge devices, and servers without a full datacenter behind it.
Canonical highlights a 1 million token context window as the agent-facing feature, saying in its Ubuntu announcement that the large window lets autonomous agents hold context across multi-step workflows that can run for extended periods — the kind of statefulness that separates a background agent from a chat box.
Deployment is handled by inference snaps, which Canonical documents as self-optimizing runtimes that detect host hardware and tune model and runtime choices for CPU, GPU, or NPU. The Inference Snaps documentation describes a standardized local API so application developers can add AI features without hardware-specific tuning. Strict confinement, automatic updates, and verified distribution are the security claims attached to the packaging.
For enterprises, Canonical frames the payoff as speed: it says the single portable artifact saves weeks of integration and validation work that teams otherwise spend standing up agent infrastructure, according to the Ubuntu announcement. Organizations can also own, fine-tune, and deploy the model wherever their agents operate, retaining control over behavior, data handling, and deployment location.
NVIDIA has been widening the open-model field this year; our coverage of NVIDIA open-sourcing Cosmos 3 world models for physical AI tracks a parallel push toward models developers can run and adapt themselves. Nemotron 3.5 Lightning extends that pattern into the agent layer on a mainstream Linux base.
The practical takeaway: a recognized open model is now a one-line install on Ubuntu, lowering the barrier for teams that want to ship agents without building inference plumbing from scratch.
