NVIDIA released Nemotron 3.5 Bolt E4B 4GB on October 23, 2026, an open small language model for local voice assistants, interactive characters, and robot interfaces that have to talk and call tools on the device. It is the smallest Bolt model, released eight days after Nemotron 3.5 Bolt 74B-A9B and Super VL, and it targets GeForce RTX, Jetson, and DGX Spark. NVIDIA designed the target workloads to fit in about 4GB, preserving headroom for other models and the application in the same pipeline. The model is post-trained from Google's Gemma 4 E4B for multi-turn tool use, instruction following, and conversation. It has a 64K context window, support for 35 languages, a conversational mode for short speech-ready replies, and a switch to turn reasoning on or off when a response has to stay fast. On the benchmarks NVIDIA selected for this workload, the company reports roughly 40 percent higher average scores than its earlier Nemotron 3 Nano 4B, and up to 80 percent fewer output tokens than the peers in that comparison. In the launch demo, a Reachy Mini on a Jetson Orin Nano with 8GB of memory uses Bolt for spoken replies and for calls into the robot's animation library, while Cosmos 3 Edge handles vision. A second example is an in-game trader that checks payment, looks up a crafting recipe, gathers ingredients, and transfers the item through tool calls. The release includes the model weights and the post-training recipe.