Install DeepSeek on Ubuntu in Three Steps
To install DeepSeek on Ubuntu, install Ollama first, then run ollama run deepseek-r1 to download the model and start using it. That is a 5.2 GB download that runs on an ordinary desktop. On a laptop with no graphics card, use ollama run deepseek-r1:1.5b at 1.1 GB instead.
DeepSeek is a free AI model you run on your own computer. Once it is downloaded you can unplug the network and it still works.
sudo. The installer changes system files.Install DeepSeek on Ubuntu
Three commands, and the third one is the model itself. Nothing here needs an account, a key or a payment method.
Step 1: Install Ollama
DeepSeek does not run on its own. Ollama is the program that downloads the model and runs it, so it has to be installed first:
curl -fsSL https://ollama.com/install.sh | shConfirm it worked. You should get a version number back:
ollama --versionThere is also a tarball install and a Docker image. Our full guide to installing Ollama on Ubuntu covers both, plus the background service and a browser chat interface.
Step 2: Download the DeepSeek Model
This pulls the 8B model, a 5.2 GB download. It takes a few minutes on a normal connection and you only do it once.
ollama pull deepseek-r1On lighter hardware, add a size tag instead, for example ollama pull deepseek-r1:1.5b. Every command below works the same way with that tag substituted. The model table further down explains which to pick.
Step 3: Start DeepSeek
This opens a chat in your terminal. Type a question, press Enter, and the answer appears below it.
ollama run deepseek-r1A working session looks like this:
Example output
An illustration of a normal session. Your wording will differ.
Type /bye to leave the chat. On anything involved, DeepSeek prints its reasoning in a think block before the answer. That is normal. Ignore it, or read it when you want to check the model’s working.
What Is DeepSeek?
What It Does
DeepSeek is an AI model you download and run on your own computer, in the same way you would install any other program. You type a question in the terminal and it answers there. It handles the usual things: explaining commands, writing and checking code, summarising text, and working through maths and logic problems.
The model in this guide is DeepSeek R1. It writes out its reasoning before giving an answer, which makes it slower than most models but better at problems with steps in them. If you want something faster for everyday questions, our guide to installing Llama on Ubuntu covers the main alternative, and the two sit side by side happily.
Who Makes It
DeepSeek is a Chinese AI lab that publishes its models openly under the MIT licence. That is one of the most permissive licences there is: you can use the models commercially and modify them without asking anyone. Nothing in this guide costs money and there is no account to create.
Update DeepSeek
Models are republished as they improve, and the tag stays the same. Pulling again fetches a newer build if there is one, and tells you it is already up to date if there is not:
ollama pull deepseek-r1That updates the model, not Ollama. Ollama itself updates separately.
First Things to Try
You do not have to open a chat every time. Put your question in quotes after the model name, and DeepSeek answers then closes. Copy any of these four and change the details to suit.
For anything you copied off a forum and do not fully trust.
Paste your own error in place of the example. This is what it is best at.
Reads the file straight off disk. Swap in your own filename.
Turns fifty lines of log into a sentence about what actually went wrong.
If you would rather have this inside your editor than in a terminal, that is a job for an editor extension. Our roundup of code editors for Ubuntu covers the ones worth wiring up.
Choose a Different DeepSeek Model
You now have the 8B model, which suits most machines. Change it only if it is too slow or it will not start. Smaller models run on weaker hardware; larger ones reason better but need far more memory.
Sizes as published by Ollama, September 2026. Bars are linear against the 671B model at 404 GB, which is why the rest barely register.
| Tag | Download | Best for | Built from |
|---|---|---|---|
deepseek-r1:1.5b |
1.1 GB | Old laptops, no graphics card | Qwen |
deepseek-r1:7b |
4.7 GB | 16 GB of RAM, no graphics card | Qwen |
deepseek-r1:8b ← default |
5.2 GB | Most desktops, or an 8 GB graphics card | Qwen3 |
deepseek-r1:14b |
9.0 GB | 16 GB graphics card. A clear step up | Qwen |
deepseek-r1:32b |
20 GB | 24 GB graphics card | Qwen |
deepseek-r1:70b |
43 GB | Workstations, 48 GB and up | Llama |
deepseek-r1:671b |
404 GB | Servers only | DeepSeek R1 |
Ollama publishes a download size for every model, and that is a fact you can check. It does not publish a memory requirement, so the Best for column is guidance based on how these sizes behave in practice. Treat it as a starting point, not a specification.
Troubleshoot DeepSeek
Error: model requires more system memory than is available
You asked for a model that will not fit. Drop a size and it will work. Going from 14B to 8B, or 8B to 1.5B, costs less capability than you would expect:
ollama run deepseek-r1:1.5bIt pauses for a long time before answering
DeepSeek writes out its reasoning before it replies, so on a slower machine there is a wait with nothing on screen. Two things help: ask shorter questions, because the thinking grows with the problem, and drop a model size. If you want instant replies for everyday questions, keep a smaller general model installed alongside and save DeepSeek for when the extra care is worth the wait.
Command not found: ollama
The install did not finish, or your terminal was open before it did. Close the terminal, open a new one, and run ollama --version again. If it still fails, re-run the install command from Step 1.
Remove DeepSeek
Models are large and easy to collect without noticing. This removes DeepSeek and frees the disk space:
ollama rm deepseek-r1To see everything you have downloaded, and what each one is using:
ollama listRemoving a model does not remove Ollama. If you want the whole thing gone, our Ollama guide has the full removal steps, including the background service and the system user the installer creates.
Frequently Asked Questions
Is DeepSeek free to use on Ubuntu?
Yes. DeepSeek publishes its models under the MIT licence, which is unusually permissive, and Ollama is open source. Running a model on your own machine costs nothing beyond disk space and electricity.
Can I run DeepSeek without a GPU?
Yes, if you keep it small. deepseek-r1:1.5b at 1.1 GB and deepseek-r1:7b at 4.7 GB run on the processor alone on most modern machines. Expect a few words per second rather than instant replies.
Which DeepSeek model should I install?
Start with deepseek-r1:8b, which is what you get from ollama run deepseek-r1. It is a 5.2 GB download and runs on an ordinary desktop. On a laptop with no graphics card use deepseek-r1:1.5b at 1.1 GB. If you have a 16 GB graphics card or better, deepseek-r1:14b at 9.0 GB is a clear step up.
What is the think block DeepSeek prints before its answer?
DeepSeek R1 works through a problem in writing before it replies, and that working appears in a think block above the answer. It is normal, not an error. You can ignore it, or read it when you want to check how the model reached its answer.
Does anything leave my machine when I run DeepSeek locally?
No. Once the model is downloaded, it runs entirely on your computer and works with the network unplugged. Ollama does list a few DeepSeek models that run on a server instead, but they have the word cloud in their name, so you will not pick one by accident.
Install Ollama, pull DeepSeek R1, and you have an AI assistant on your own machine that never sends a word anywhere.
Related guides: Install Ollama on Ubuntu · Install Llama on Ubuntu · Best Open Source AI Tools for Ubuntu