NVIDIA’s Nemotron 3.5 Bolt E4B Fits in 4 GB and Tops a Multi-Turn Test

A 4 GB model built to talk and use tools on small devices.

@NVIDIAAI released Nemotron 3.5 Bolt E4B 4GB, its newest small open model, for local assistants, interactive characters and robots that carry on a conversation and use tools. Its demo shows a Reachy Mini robot chatting and moving with Bolt handling the conversation. Weights, data and recipes are open on Hugging Face.

@ScaleAILabs ranks it the top small language model on MultiChallenge at 56.03%, 12.8 points ahead of Gemma 4 E4B at 43.2%. The test checks how models hold up across realistic multi-turn conversations. NVIDIA’s RTX account says it runs locally on GeForce RTX, Jetson, RTX Spark and DGX Spark.