NVIDIA Nemotron 3.5 Lightning

NVIDIA Nemotron 3.5 Lightning is now available on Ollama and runs completely on a user's own device, according to the Ollama post dated August 11, 2026. The model is a 30 billion parameter open model from NVIDIA with 3B active parameters, described as built for agents that stay running: gathering context, calling tools, and working through multi-step tasks.

Ollama said the model targets agentic work such as reading a file, calling a tool, sorting a result, and retrying a failed step, and that at 3B active parameters per token it is built for local systems rather than the datacenter. It uses a hybrid Mixture-of-Experts architecture. Ollama lists supported local hardware as NVIDIA RTX PCs, NVIDIA RTX PRO workstations, NVIDIA DGX Spark and DGX Station, plus the datacenter and cloud. The model has a context length up to 1M tokens, which Ollama says leaves room for long tool histories across multi-turn workflows.

Ollama said the model was developed with the Nemotron Coalition and trained for coding, tool calling, instruction following and multi-turn work. It supports speculative decoding using multi-token prediction (MTP), DFlash or DSpark, which Ollama says offers up to 4x higher throughput than comparable open models. Ollama states the model offers 4x higher throughput and 30% faster task completion time compared to other leading open models of similar size, with full results and test configurations in NVIDIA's launch blog.

Ollama lists example workloads including long-running personal assistants, coding sub-agents, security operations, a local tier alongside a cloud model, and post-training a specialist. The model is open weights and trained on open datasets. Ollama lists run commands for general chat, Claude Code, OpenClaw, Hermes Agent and OpenCode, and notes an MLX variant, nemotron-3.5-lightning:30b-mlx, for Apple silicon.