# Nemotron 3.5 Bolt E4B

> Canonical: https://www.overmindlab.ai/models/nvidia-nemotron-3-5-bolt-e4b-bf-16

8B Gemma 4 finetune with thinking that can be turned off.

| Property | Value |
| --- | --- |
| Model ID | nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16 |
| Provider | NVIDIA |
| Group | Nemotron |
| Parameters | E4B (~8B) |
| Tier | Small |
| Context window | 64K |
| Max training context | 64K |
| Tool calling | Yes |
| Training methods | LoRA |
| Training cost | from $0.46 |
| Serving cost (per 1M output tokens) | from $4.00 |

## Overview

42 layers, a 65,536-token window, and tool calling. LoRA is the enabled training method.

Nemotron 3.5 Bolt E4B is an NVIDIA finetune of Gemma 4 E4B. It trains with LoRA on the Gemma 4 image and serves on the pinned vLLM image.

## Good for

- Agents that start from the Bolt E4B checkpoint
- LoRA on a Gemma 4 E4B architecture
- Tool-calling rows

Train this model on your agent's data: https://docs.overmindlab.ai/models/training.md