Nemotron 3.5 Lightning

Nemotron 3.5 Lightning is NVIDIA's small, fast open model, and it suits developers who want an agent model on a single GPU. The NVIDIA model card gives 30B total parameters with 3B active, a context of up to 1M tokens, and a release date of 11 August 2026. It uses a hybrid of Mamba-2, mixture-of-experts and attention layers, and the licence is OpenMDW 1.1. NVIDIA says it runs on one H100 or A100 with 80GB. OpenRouter lists it with a 262,144-token window, text only, at $0.07 per million input tokens and $0.20 per million output tokens, so hosted access is shorter than the card's maximum. That gap is the catch: check which context your provider serves before you plan around a million tokens. Figures come from the NVIDIA card and the OpenRouter model list. We have not run it ourselves. We would use it for high-volume agent steps where speed and cost matter more than peak quality, and keep a larger model for the hard parts.