NVIDIA Nemotron 3.5 Bolt E4B: developer tools for small teams
Official site link, not an affiliate link. We do not have an approved referral link for NVIDIA Nemotron 3.5 Bolt E4B.
NVIDIA Nemotron 3.5 Bolt E4B helps you build software by starting with a small open model from NVIDIA for fast local voice and device assistants. It is best for developers building local voice assistants, game characters or on-device agents. People also use it to work with voice and audio. A free plan is available.
What it does
NVIDIA’s model card describes an 8B-parameter model (4B effective) post-trained from Gemma-4-E4B-it, with improved multi-turn tool use, instruction following and input grounding, a context of up to 64K tokens, and reasoning that can be switched on or off. It is text in, text out. NVIDIA publishes a full-precision BF16 release for customization and a Q4_K_M “4GB” build for llama.cpp, plus a DFlash drafter for faster speculative decoding.
Official site:build.nvidia.com/nvidia/nemotron-3.5-bolt-e4b.
Editorial note
Nemotron 3.5 Bolt E4B is a small open language model that NVIDIA post-trained from Google’s Gemma 4 E4B. NVIDIA says it is built for fast local voice and device assistants, and a 4GB quantized build is available for local use.
Who it’s for
- For: Developers building local voice assistants, game characters or on-device agents.
- Not ideal for: people who want a ready-made app; it is a model for developers.
Key use cases
- Build apps & code
- Voice & audio
What you can do with it
- Free access: NVIDIA publishes the weights on Hugging Face under the OpenMDW-1.1 license and lists a free API endpoint for prototyping. This matters when you want to test before you pay.
- NVIDIA API plan: Hosted endpoint on build.nvidia.com for trying the model. This matters when the free tier is not enough.
- Help documentation: Official pricing and product pages at build.nvidia.com/nvidia/nemotron-3.5-bolt-e4b. This matters when you set it up yourself.
How to use NVIDIA Nemotron 3.5 Bolt E4B: step-by-step tutorial
Five to eight steps from the account to a first useful result. Where the official guide names a control, the step uses that name. Where it does not, the step stays general.
- Open the model page on Hugging Face (NVIDIA-Nemotron-3.5-Bolt-E4B-4GB).
- Download the Q4_K_M GGUF file named in the quick start.
- Load it in llama.cpp, optionally with the DFlash drafter for speed.
- Test a multi-turn task with a tool call, the use the model card highlights.
- Or try the free hosted endpoint on build.nvidia.com first.
These steps follow the official guide: Model card (4GB build).
Frequently asked questions
Is Nemotron 3.5 Bolt free?
The weights are a free download under the OpenMDW-1.1 license, and NVIDIA says the model is ready for commercial or non-commercial use, checked October 24, 2026.
What is it based on?
NVIDIA says it is post-trained from Google’s Gemma-4-E4B-it.
How big is it?
8B total parameters with 4B effective, per the model card, with a 4GB quantized build for local use.
Does it handle speech?
No. The model is text in and text out; NVIDIA positions it as the language model inside local voice assistants.
Pricing
Prices checked on the official pricing page.
- Open weights
-
Free download
BF16 and a 4GB Q4_K_M build on Hugging Face, OpenMDW-1.1 license, commercial or non-commercial use
- NVIDIA API
-
Free prototyping endpoint
Hosted endpoint on build.nvidia.com for trying the model
- Limits: Text only; you need your own speech and app layers for a voice assistant.
- Plans change: vendors update limits and prices; confirm on the official page before you buy.
Facts
- Access
- Ongoing free plan
- Paid plans from
- Free
- Checked
- Category
- Category: Development tools (31 in the library).
- On the record
- NVIDIA publishes the weights on Hugging Face under the OpenMDW-1.1 license and lists a free API endpoint for prototyping.