NVIDIA Nemotron 3.5 Bolt: a small open model for local voice assistants
Published· By Mishal Brands Editorial · ai-news, ai-tools
NVIDIA reports Nemotron 3.5 Bolt E4B is an 8B-parameter open model (4B effective) for fast local voice and device assistants, with a free 4GB build on Hugging Face.
NVIDIA published Nemotron 3.5 Bolt E4B, a small open language model it says is built for fast local voice and device assistants (model card).
What's in it, as reported
- Size. 8B parameters, 4B effective, post-trained from Google's Gemma-4-E4B-it.
- Skills. NVIDIA cites better multi-turn tool use, instruction following and input grounding.
- Context. Up to 64K tokens, with reasoning you can switch on or off. Text in, text out.
- Builds. A full-precision BF16 release for customizing, and a 4GB Q4_K_M build for llama.cpp (BF16 model card).
Cost and license
The weights are free under the OpenMDW-1.1 license, which NVIDIA says allows commercial or non-commercial use. NVIDIA also lists a free API endpoint for prototyping (NVIDIA API catalog).
How to try it
- Try it in the browser on the NVIDIA API catalog.
- To run it locally, download the 4GB build from Hugging Face and load it in llama.cpp.
- Test it with a short tool-use task, like "check the weather, then set a reminder".
More: Nemotron Bolt tool page.
Tools in this post
- NVIDIA Nemotron 3.5 Bolt E4B: A small open model from NVIDIA for fast local voice and device assistants.