What is NVIDIA Nemotron 3.5 Bolt E4B 4GB?

NVIDIA Nemotron 3.5 Bolt E4B 4GB is a downloadable text model post-trained from Gemma-4-E4B-it. It targets conversation, instruction following, and multi-turn tool use.

Developers can use it for local assistants, interactive personas, and embedded interfaces. Jetson AI Lab provides a deployment path using a llama.cpp container, with a locally served API for connecting applications.

Key features

Quantized downloadable checkpoint

The Q4_K_M release has 8B total parameters and 4B effective parameters.

Reasoning controls

Chat-template settings enable or disable thinking and conversational mode; a runtime thinking budget is also documented.

Extended text context

The model documents a context length of up to 64K tokens.

Jetson deployment guidance

The deployment selector covers listed Orin and Thor modules and provides server and client instructions.

Who is it for?

Device assistant development

Prototype a conversational interface for a Jetson-based device.

Interactive character design

Explore dialogue and instruction following for an interactive persona.

Tool integration testing

Evaluate a local conversational component within an application that exposes tools.

How to use NVIDIA Nemotron 3.5 Bolt E4B 4GB

Review the model card and choose the Q4_K_M checkpoint.
For Jetson, select the module in Jetson AI Lab and copy its llama.cpp container command.
Start the model server using the documented configuration.
Connect through the documented chat-completion API and evaluate representative conversations.

NVIDIA Nemotron 3.5 Bolt E4B 4GB pricing

Free

Model materials are available without a license fee under OpenMDW-1.1, subject to its terms. Running the model requires your own compute or hosting.

Check current pricing ↗

What to know before you use it

The 4GB name describes approximate memory use in selected workloads, not a universal requirement. NVIDIA reports configurations exceeding 4GB; context length, hardware, and runtime affect consumption.
Inputs and outputs are text. A voice assistant requires separate audio components.
Reasoning and conversational controls depend on the serving runtime applying the chat template.
Jetson deployment requires selecting a compatible module and running the documented container setup.

Frequently asked questions

Can it be used commercially?

Yes. NVIDIA permits commercial and non-commercial use under the applicable license terms.

How does an application connect on Jetson?

Start the llama.cpp server, then send chat-completion requests using the documented local or LAN API connection.