Ollama users can now run NVIDIA Nemotron 3.5 Lightning, an open model built for agents that stay active across long tasks. The model has 30 billion total parameters, with about 3 billion active during inference, which makes it lighter to run than its full size might suggest.
The release is aimed at workflows where an assistant gathers context, calls tools, and works through multiple steps on a user’s own hardware. That is different from a one-shot chatbot exchange. Persistent agents need speed, predictable cost, and enough capability to keep state across a task without sending everything to a hosted service.
Availability through Ollama lowers the setup barrier for developers and local-AI users who already rely on the platform for model serving. The main limitation is that efficient local models still trade off against the strongest cloud systems on difficult reasoning and broad knowledge. But for private, repeatable, tool-heavy work, a fast open model can be more useful than a larger remote model that is slower, more expensive, or harder to control.