Hermes-3-Llama-3.2-3B

Hermes-3-Llama-3.2-3B

A 3.2B chat and assistant model that follows instructions, calls functions, writes structured output and code. For developers and hobbyists who run a model on their own computer.

Made by NousResearch on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Hermes-3-Llama-3.2-3B is a 3.2B general-purpose language model from Nous Research. It starts from the Llama-3.2 3B base model, with every weight updated. It handles multi-turn chat, reasoning, role-play, code, function calling and structured output. The card calls it a step up from Hermes 2.

Who it suits: It suits developers and hobbyists who want a small assistant model for their own computer, including one that can call functions in an agent setup.

Industry | Who uses it | How they use it | Example
Software | Developers | Build assistants that call functions and return structured output | Connect it to a tool-calling workflow
Education | Students | Chat and work through reasoning questions | Ask a multi-step question, then follow up
Creative | Writers and game makers | Run role-play and story chat | Hold a long conversation in character
Research | Hobbyists | Test a small fine-tune on modest hardware | Compare it with other 3B models

If you need languages other than English, you can skip it, since the card lists English only.

Where to use it: You run it in Python with the transformers library, on Windows, macOS or Linux. Hugging Face also lists 30 quantized copies that other people published.

• Python with the transformers library
• Windows, macOS or Linux with the huggingface_hub package
• Agent and chat apps that you build around the model

Use case | Who | What they get back
Assistant chat | Everyday users | Multi-turn answers that keep track of the conversation
Tool calling | Developers | Function calls and structured output
Code help | Programmers | Generated code from a plain request

How to set it up: You need Python, the transformers library and the Hugging Face hub package. Hugging Face estimates about 6.4 GB for the weights in BF16 and about 1.6 GB at 4-bit. The card text available here has no run instructions.

• Python on Windows, macOS or Linux
• The transformers library
• The Hugging Face hub package

1. Install the Hugging Face hub package.
2. Download the model files.
3. Load them with the transformers library.

pip install -U huggingface_hub
hf download NousResearch/Hermes-3-Llama-3.2-3B

How to use it: Load the model with transformers, send it your chat messages and read its reply. The card text available here shows no usage code, so follow the library's usual way of loading a chat model.

Example: You ask it a question and tell it to answer in a fixed structure. The model replies in that structure, which your own program can then read.

For newcomers and experts: If you are new to this, think of it as a chat assistant small enough for a home computer. A first small step is to download it and ask a few questions that build on each other.

If you know the field, it is a full-parameter fine-tune of Llama-3.2 3B that builds on Hermes 2, adding better function calling and structured output. The authors place it close to Llama-3.1 Instruct models on general tasks, with strengths and weaknesses on each side.

Chat messages go in
The Llama-3.2 3B fine-tune reads the conversation
It reasons, calls functions or writes code
A reply or structured output comes back

Good to know: The card reports an average of 64.00 on its GPT4All set and 34.36 on AGIEval. The chart shows GPT4All accuracy by task.

Title: GPT4All accuracy by task, 0-shot
boolq: 0.8327
piqa: 0.7639
arc_easy: 0.7399
winogrande: 0.6590
hellaswag: 0.5453
arc_challenge: 0.4411
openbookqa: 0.3480

• License: Llama 3 license
• Cost to run: no API key or paid service is named, and the weights are about 6.4 GB in BF16
• Data it reaches: the download comes from Hugging Face, and the model works on the messages you give it
• Not for: work in languages other than English

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 3.2B
• Weights: about 6.4 GB in BF16; about 1.6 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 4,602
• Likes: 187
• License: llama3
• Languages: en
• Based on: meta-llama/Meta-Llama-3.2-3B
• Quantized versions: 30
• Fine-tunes: 19
• Tags: transformers, safetensors, llama, text-generation, llama-3, instruct, finetune, chatml, gpt4, distillation, axolotl, roleplaying

Links

• Model: https://huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B
• Files: https://huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download NousResearch/Hermes-3-Llama-3.2-3B

Then follow the usage in its model card. Nothing in this kit runs by itself.