NVIDIA's Nemotron 3.5 Lightning is an open ~30B-parameter MoE model released Aug 11, 2026, purpose-built for high-volume agent execution. It activates only ~3B parameters per token, delivers up to 4x output throughput versus similarly sized models, and runs a 1M-token context on a single H100 or DGX Spark. NVIDIA ships it with the NeMo Switchyard routing library for multi-agent workflows.
Parameters
30B (MoE, ~3B active)
Context Window
1M
License
NVIDIA Open Model License
Release Date
2026-08-11
API Pricing
API pricing for this model is not yet available
Strengths
- •Up to 4x output throughput vs similar-sized models for agent execution
- •1M-token context window
- •Runs on a single H100 / DGX Spark
- •NeMo Switchyard routing for multi-agent workflows
Weaknesses
- •Reasoning quality may trail larger frontier models
- •Tuned for speed/cost over raw generation quality
- •Tied to NVIDIA's NeMo ecosystem
Use Cases
- •Tool calls, validation and sub-agent steps in agent workflows
- •Cybersecurity alert triage
- •Financial risk monitoring
- •Customer-service automation