Install Llama on Ubuntu, Without Sending a Word to the Cloud
To install Llama on Ubuntu, install Ollama first, then run ollama run llama4:16x17b to download Llama 4 and start chatting. Scout is a 67 GB download, so if your machine cannot take that, use ollama run llama3.1:8b at 4.9 GB instead.
Llama is a family of AI models built by Meta. Ollama is the free tool that downloads and runs them. Two different things with confusingly similar names: you install Ollama once, then use it to run Llama, Mistral, Qwen or anything else.
sudo. The installer changes system files.What Llama 4 Is, in Short
Llama 4 is built differently from every Llama before it. That changes what it can do, and it changes how much room it needs.
- It is split into many small sections, and only a few switch on per question. That is what keeps it fast.
- The whole model still sits on your drive, which is why it answers like a mid-sized model but stores like a huge one.
- It reads images as well as text. Older Llama versions are text only.
- It is the most capable version you can run at home, and by far the biggest download.
Download Sizes, Drawn to Scale
Below are the download sizes reported by Ollama. The bars are drawn to scale, which is the fastest way to see the problem.
Sizes as published by Ollama, September 2026. Bars are linear against Maverick at 245 GB, which is why the small models barely register.
Scout is the entry point to Llama 4 and the version this guide installs. It is a serious download, so check your free disk space before you start it. If your machine cannot take 67 GB, Llama 3.1 8B is the lighter route and the rest of this guide works the same way with that tag substituted.
Which Llama Model Should You Run?
Find the card that matches your machine. If you are between two, take the smaller one first. Downloading a bigger model later is easy. Waiting an hour for one that will not fit is not.
67 GB download. The newest Llama. It reads images as well as text. Needs a workstation with lots of memory, or more than one graphics card.
4.9 GB download. The lighter route if Llama 4 is out of reach. Handles code review, long writing tasks and general questions without feeling limited.
43 GB download. Noticeably stronger reasoning than 8B. Needs either a large GPU or enough system RAM to spill into, which slows it down.
2.0 GB download. Runs on CPU alone. Good for summarising, rewriting and simple questions. Drop to llama3.2:1b if it still struggles.
The graphics card sizes above are our guidance, based on how big each model is. Ollama publishes download sizes, not graphics memory requirements, so treat these as a starting point rather than a hard rule.
Install Ollama First
Llama does not run on its own. Ollama is the program that downloads the model files and runs them, and it has to be installed before anything on this page works. One command does it:
curl -fsSL https://ollama.com/install.sh | shThat script is the quickest route. There are also a tarball install and a Docker image, plus the background service, updating and clean removal. Our full guide to installing Ollama on Ubuntu covers all of it. Everything below assumes Ollama is working.
What a Working Install Looks Like
Whichever method you picked, this is where you are heading. The first time you run a model, Ollama downloads it and shows a progress bar. After that it drops you straight into a chat.
Example output
An illustration of a normal session. Your download ID and wording will differ.
Download and Run Llama
Ollama downloads a model the first time you ask for it. You do not need a separate download step, though pulling first is useful when you want it finished before you sit down.
This fetches Llama 4 Scout, which is 67 GB. Start it when you do not need the connection for anything else. Running lighter hardware? Swap the tag for llama3.1:8b and every command below still works.
ollama pull llama4:16x17bThis drops you into an interactive chat. Type a question, press Enter, and the model answers in your terminal.
ollama run llama4:16x17bType /bye to leave the chat. Running ollama on its own opens an interactive menu if you would rather pick from a list.
Models add up quickly. These two commands show what is on disk and remove what is not earning its space.
ollama list
ollama rm llama4:16x17bWhen Llama Will Not Run
Most install problems belong to Ollama rather than Llama, and our Ollama guide covers those. This is the one you are most likely to hit with Llama itself.
Error: model requires more system memory than is available
You asked for a model that will not fit. This is the most common wall people hit with Llama 4, because Scout needs far more than a typical desktop has. Drop to a smaller model and it will work:
ollama run llama3.1:8bFrequently Asked Questions
Is Llama free to use on Ubuntu?
Yes. The model weights are free to download and run under Meta’s community licence, and Ollama is open source. You pay nothing to run a model locally. The only cost is disk space, electricity and your own hardware.
Can I run Llama without a GPU?
Yes, but keep the model small. Llama 3.2 1B and 3B run on CPU alone on most modern machines. Expect a few words per second rather than instant replies. Anything at 8B or above without a GPU becomes slow enough to be frustrating.
Which Llama model should I run on Ubuntu?
Llama 4 Scout, tagged llama4:16x17b, is the current release and the one to run if your hardware allows it. It is a 67 GB download and expects a high-memory workstation or multiple GPUs. If that is out of reach, llama3.1:8b at 4.9 GB is the standard lighter choice, and llama3.2:3b at 2.0 GB runs without a GPU at all.
Can I run Llama 4 on a normal desktop?
Usually not. Llama 4 splits itself into many sections and only runs a few at a time, but the whole model still has to be stored, so Scout is a 67 GB download and Maverick is 245 GB. Check your free disk space first. If that is out of reach, Llama 3.1 8B gives you most of the everyday capability at 4.9 GB.
What is the difference between Llama and Ollama?
Llama is the AI model, built by Meta. Ollama is the free program that downloads models and runs them on your computer. The names look alike but they are separate projects. You install Ollama once, then use it to run Llama, Mistral, Qwen or any other model.
Install Ollama, pull Llama 4, and you have a private assistant that works with the network unplugged.
More Ubuntu guides: Install Ollama on Ubuntu · Best Open Source AI Tools for Ubuntu · Install Docker on Ubuntu · Excellent Code Editors for Ubuntu