By Local AI Scout · Models · Local · Guide · Docs-only · Checked November 18, 2026
Short answer: You can run DeepSeek-R1 on your own computer with Ollama in one command: ollama run deepseek-r1. That default tag is the 8B version, a 5.2 GB download. Smaller and larger versions run from a 1.1 GB model up to the full 671B model at 404 GB, which is far beyond a normal computer. Most of the sizes you'd run at home are "distilled" models: smaller Qwen or Llama models trained on DeepSeek-R1's reasoning. Ollama's page lists the weights under the MIT License, with the Llama-based versions under Llama's own licenses. Run locally, your prompts stay on your machine. This is a documentation review, not a benchmark we ran.
This guide follows one reader: a bookkeeper who wants a reasoning model for checking spreadsheets, without sending client numbers to a cloud service.
The sizes, from Ollama's library
From Ollama's DeepSeek-R1 page (opened November 18, 2026):
| Tag | Download | Based on |
|---|---|---|
deepseek-r1:1.5b | 1.1 GB | Qwen |
deepseek-r1:7b | 4.7 GB | Qwen |
deepseek-r1:8b (the default) | 5.2 GB | Qwen3 |
deepseek-r1:14b | 9.0 GB | Qwen |
deepseek-r1:32b | 20 GB | Qwen |
deepseek-r1:70b | 43 GB | Llama |
deepseek-r1:671b | 404 GB | full DeepSeek-R1 |
Ollama lists a 128K context window for the distilled sizes and 160K for the full model.
Which size for your computer
The download size is a rough floor for the memory you'll need, and you need headroom on top ‡. Our memory guide explains the sums. As a starting point ‡:
- 8 GB of memory: try
1.5b. - 16 GB:
7bor8b. - 32 GB:
14b, and possibly32bon a Mac with unified memory. - 64 GB or more:
32bcomfortably;70bneeds more.
Our bookkeeper has a 16 GB laptop, so she starts with the default.
Run it
- Install Ollama; our Ollama install guide has the steps.
- Open a terminal and run
ollama run deepseek-r1(or add a tag, such asollama run deepseek-r1:1.5b). - Ask a question. R1 is a reasoning model, so it usually shows its thinking before the answer ‡. That's slower, but useful for checking its work.
No terminal? Desktop apps such as LM Studio can download and run models too.
Licenses and business use
Ollama's page says the DeepSeek-R1 weights are under the MIT License, which allows commercial use and modification. The Qwen-based versions come from Qwen 2.5, which Ollama lists as Apache 2.0. The 70b version is Llama-based and falls under Llama 3.1 and 3.3 licenses. Read can I use local AI models commercially? before you build it into client work.
Local versus DeepSeek's own app
Running the model through Ollama keeps your prompts on your computer. Using DeepSeek's own website or app is a cloud service with its own terms ‡. For client data, local is the point.
Compare with other models: run Qwen locally, run Llama locally, or browse our models hub. On a Chromebook? See local AI on a Chromebook.
Frequently asked questions
Can I run DeepSeek locally?
Yes. With Ollama installed, run ollama run deepseek-r1, which downloads the default 8B version at 5.2 GB. Smaller and larger sizes are available, from 1.1 GB up to the full model at 404 GB.
What is the smallest DeepSeek-R1 model?
On Ollama, deepseek-r1:1.5b is a 1.1 GB download. It's a distilled model based on Qwen, small enough to try on a computer with 8 GB of memory, though answers are weaker than the larger sizes.
Is DeepSeek-R1 free for commercial use?
Ollama's page lists the DeepSeek-R1 weights under the MIT License, which allows commercial use. The Qwen-based versions are Apache 2.0, and the Llama-based 70b version follows Llama 3.1 and 3.3 licenses, so check the license for the size you use.
Sources
- Ollama, "deepseek-r1" library page: https://ollama.com/library/deepseek-r1 (opened November 18, 2026)
- Ollama, "Download Ollama on Linux": https://ollama.com/download/linux (opened November 18, 2026)
How we checked: a documentation review of Ollama's pages, opened November 18, 2026. We haven't benchmarked these models for this guide. ‡ marks something not stated on an official page we opened, or our own reading. See our editorial policy.