NVIDIA's new 3B model just turned coffee stains into planets
Nemotron 3.5 Lightning dropped today: a 30B mixture-of-experts model with only 3B active parameters, a 1M-token context window, and open weights. NVIDIA claims up to 4x faster output than competing models in its class — light enough to run on a single GPU!
We put it through our napkin benchmark: the model gets a photo of a napkin with three coffee ring stains and one prompt — turn the stains into an animated drawing, one self-contained HTML file.
Output: 9,600 tokens
The model read the position and size of every stain and drew a planet exactly over each ring — one with orbiting moons, one with an asteroid field, one big and glowing. Then it added a pencil that sketches them in, stroke by stroke. No missed rings, no misaligned circles — the stains became the planets
Nemotron 3.5 Lightning is already live on aimlapi{.}com for FREE
Introducing NVIDIA Nemotron 3.5 Lightning⚡
An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volume, specialized tasks faster.
It delivers up to 4x the output speed of similar-sized models.