Nemotron 3.5 Bolt E4B

Name: Nemotron 3.5 Bolt E4B
Author: NVIDIA

NVIDIA · Nemotron 3.5 · released 2026-10-05 · data checked 2026-11-01

Nemotron 3.5 Bolt E4B is a 8B-parameter dense model from NVIDIA (United States), released 2026-10-05 under the OpenMDW-1.1 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 5.6 GB of memory at 4-bit with an 8K-token context.

Small LM Index

—

Needs 2 core benchmarks (has 0)

Parameters

4.5B effective

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

65K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best small model on that benchmark. Sources: 3 reported by NVIDIA. How we label sources.

τ²-Bench Lab-reported Agentic55.1

Class best: Qwen3.5 9B 79.1

IFBench Lab-reported Instruction following63.8

Class best: Granite 4.2 8B 79.3

BFCL Lab-reported Agentic · BFCL v368.5

Class best: Hunyuan 7B Instruct 70.8

Specifications

Lab: NVIDIA (US · United States)
Based on: Gemma 4 E4B (fine-tune or distillation)
Architecture: Dense transformer
Total parameters: 8B
Active parameters: All (dense)
Size class: Small
Context window: 65K tokens
Input modalities: text, image, audio
Reasoning: Hybrid (thinking on/off)
License: OpenMDW-1.1
Openness: Open weights
Memory to run: ~5.6 GB at 4-bit, 8K context (more sizes)
Hugging Face: 12.7K downloads (30 days) · 64 likes

Weights on Hugging Face ↗

Compare with

Nemotron 3.5 Bolt E4B vs Granite 4.2 8B (8B, index 63.5)
Nemotron 3.5 Bolt E4B vs Ministral 3 8B (8B, index 55.2)
Nemotron 3.5 Bolt E4B vs Gemma 4 E4B (8B, index 38.2)

Sources

https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16

Something wrong here? Report it.

Run it locally

Uses nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB (Q4_K_M). Swap the tag for another size: Q4_K_M.

Ollama

ollama run hf.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB:Q4_K_M

llama.cpp

llama-server -hf nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/NVIDIA-Nemotron-3.5-Bolt-E4B-OptiQ-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	65K context
16-bit F16	17 GB	17 GB	17 GB
8-bit Q8_0	9.2 GB	9.5 GB	9.7 GB
4-bit Q4_K_M	5.6 GB	5.8 GB	6.1 GB

Check it against your GPU, Mac or phone →