NVIDIA Nemotron 3.5 Lightning Now Available in AWS SageMaker JumpStart

NVIDIA's open Nemotron 3.5 Lightning — a 30B-parameter Mixture-of-Experts model with 3B active — lands in Amazon SageMaker JumpStart, built for high-volume agentic workloads that don't need frontier-scale compute.

NVIDIA's lightweight Nemotron 3.5 Lightning model is now available in Amazon SageMaker JumpStart, letting teams deploy the open model without configuring their own serving infrastructure. AWS announced the launch Monday, positioning the model for the high-volume end of agent workflows.

The model uses a hybrid Mixture-of-Experts architecture with 30 billion total parameters but only 3 billion active per forward pass, so it runs on a single GPU. NVIDIA reports up to 4x higher throughput and up to 30% faster task completion on repetitive agentic work — the classifying, extracting, and policy-checking steps that don't need a frontier model. A 1M-token context window lets an agent carry state across long-running sessions, and DFlash speculative decoding trims per-token latency.

Lightning is distilled from NVIDIA's frontier Nemotron 3 Ultra and developed with the Nemotron Coalition, trained specifically for tool use across popular agent harnesses. NVIDIA positions it as the high-volume tier in a system-of-models approach, with NeMo Switchyard routing each workflow step to the best-suited model.

Sources: AWS Machine Learning Blog | NVIDIA Blog