Hermes 3 Llama-3.1 8B

Hermes 3 Llama-3.1 8B

An 8B general-purpose chat model from Nous Research, built on Llama 3.1, with function calling, structured output, role play and code help. It suits developers who want a steerable open assistant they can run themselves.

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-3.1 8B is a general-purpose language model from Nous Research, built on Meta's Llama 3.1 8B. It is the newest model in the Hermes line. Compared with Hermes 2 it adds better agent behavior, role play, reasoning, multi-turn chat and coherence over long contexts. It also handles function calling and structured output more reliably, and writes better code. The Hermes line aims to give the user control through system prompts instead of fixed rules from the maker.

Who it suits: It suits developers and hobbyists who want an assistant they can steer with a system prompt and run on their own hardware. It also suits people who build agents that call functions or return structured data.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Software | Developers | Add a local chat assistant to an app | Serve it behind a chat endpoint in the same format as OpenAI's
Agents | Agent builders | Use function calling and structured output | Have it return a JSON call for a tool
Creative | Writers and game makers | Set a role with a system prompt | Run a character chat with a custom persona

If you need the strongest model available, you can skip it. The card says it is competitive with Llama 3.1 Instruct, with different strengths and weaknesses.

Where to use it: The record lists the transformers library, so Python is the main place. The prompt format matches the one OpenAI's chat endpoints use, so many chat tools can talk to it.

• Python with the transformers library
• Chat servers and tools that accept the ChatML format
• Windows, macOS or Linux with Python, to download the files

Title: Jobs it does
Use case | Who | What they get back
General assistant chat | Product teams | Replies that follow your system prompt
Function calling | Agent builders | Calls built from the function signatures you supply
Code help | Developers | Code drafts and explanations

How to set it up: You need Python with pip and the transformers library, plus memory for about 16.1 GB of BF16 weights, or about 4 GB at 4-bit by estimate. You also load the model and tokenizer, as with any transformers model.

• Python with pip
• The transformers library
• Memory for about 16.1 GB of BF16 weights, or about 4 GB at 4-bit

1. Install the Hugging Face download tool.
2. Download the model files to your computer.
3. Load the model and tokenizer with transformers.

pip install -U huggingface_hub
hf download NousResearch/Hermes-3-Llama-3.1-8B

How to use it: The model uses the ChatML prompt format, where special tokens mark the start and end of each turn and its role. The card says the tokenizer has a chat template, so you can build the prompt with apply_chat_template. Set the option that adds a generation prompt, so the model continues as the assistant. You can drop the system message if you do not want one. For function calling, the card says to use a system message with function signatures as JSON.

messages = [
    {"role": "system", "content": "You are Hermes 3."},
    {"role": "user", "content": "Hello, who are you?"}
]
gen_input = tokenizer.apply_chat_template(messages, return_tensors="pt")
model.generate(**gen_input)

Example: You load the model, set a system message that names a role, and ask it to greet you. The chat template turns the two messages into the ChatML format, and the model answers as the assistant. Change the system message to change the tone and rules.

For newcomers and experts: If you are new, this is a chat model you can run yourself, and the system message is how you tell it who to be. Start with a short system message and one question.

If you know the field, ChatML gives structured turns and the same format as OpenAI's chat endpoints, which makes it easy to swap into existing clients. The card says the Hermes 3 series builds on Hermes 2 with more reliable function calling and structured output. It compares benchmark results with Llama 3.1 Instruct in a chart image.

A system message and a user message go in
The chat template builds a ChatML prompt
The model generates the assistant turn
You read a reply, or a function call as JSON

Good to know: The model is small enough for a single good GPU, and smaller still when quantized. People have shared 52 quantized versions and 40 fine-tunes.

Title: Approximate weight size
BF16: 16.1 GB
4-bit estimate: 4 GB

• License: Llama 3 community license
• Cost to run: none named
• Data it reaches: the model files from Hugging Face, and the prompts you give it
• Not for: uses that need a verified answer without a check, or anything that breaks the Llama 3 license terms

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 8B
• Weights: about 16.1 GB in BF16; about 4 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 336,953
• Likes: 550
• License: llama3
• Languages: en
• Based on: meta-llama/Meta-Llama-3.1-8B
• Quantized versions: 52
• Fine-tunes: 40
• Tags: transformers, safetensors, llama, text-generation, llama-3, instruct, finetune, chatml, gpt4, distillation, axolotl, roleplaying

Links

• Model: https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B
• Files: https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B/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-3.1-8B

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