Ollama v0.32.9 adds NVIDIA Nemotron 3.5 Lightning with 3B active params

Ollama's v0.32.9 release brings native support for NVIDIA Nemotron 3.5 Lightning, a 30B mixture-of-experts (MoE) model that activates only 3B parameters per token. This architecture dramatically reduces compute and memory overhead while retaining strong reasoning capabilities, making it ideal for persistent, always-on AI agents that need low-latency responses. The model is backed by NVIDIA's NemoClaw open-source security and management stack, and integrates with agent harnesses like OpenClaw and Hermes Agent.

The update also introduces the Nemotron 3 architecture to Ollama's model router and fixes a boundary condition in the Muse Glimmer function-calling parser, improving tool-calling reliability. Developers can immediately pull the model with 'ollama run nemotron-3.5-lightning'. For professionals building autonomous agents, this means running an efficient, open-weight model locally or on edge infrastructure without high-end GPU clusters—a significant step toward practical, deployable agent workloads.