Nemotron 3.5 Bolt E4B

NVIDIA's open small model for local voice assistants, characters, and device agents.

Also known as Nemotron 3.5 Bolt, NVIDIA-Nemotron-3.5-Bolt-E4B

What it does

Nemotron 3.5 Bolt E4B is NVIDIA's open small language model for conversational agents, released October 23, 2026. NVIDIA describes a Nemotron post-train of Gemma 4 E4B for voice and device assistants, interactive characters, and robotic or IoT interfaces on GeForce RTX, Jetson, RTX Spark, and DGX Spark. The research post says target workloads fit in about 4GB, that average quality is about 40% higher than its prior model on key conversational-agent benchmarks, and that it outputs up to about 80% fewer tokens than peers. It lists multi-turn tool use, a conversational mode, configurable reasoning, a 64k context, and 35 languages. The BF16 model card lists 8 billion total parameters and 4 billion effective, text in and out, up to 64K tokens of input and of output, and the OpenMDW-1.1 license. Gemma 4 use is under Apache 2.0. Neither page lists a hosted price.

Who is Nemotron 3.5 Bolt E4B for?

People running a local voice assistant, an in-game character, or a device agent on NVIDIA hardware.

Tags

• open-weight
• local
• agents