How to Run DeepSeek R1 Locally on Windows, macOS, Android & iPhone

The DeepSeek R1 model from a Chinese team has rocked the AI industry. It has overtaken ChatGPT and achieved the top position on the US App Store. Not just that, DeepSeek has rattled the US tech stock market with its groundbreaking R1 model, which claims to match ChatGPT o1. While you can access DeepSeek R1 for free on its official website, many users have privacy concerns as the data is stored in China. So, if you want to run DeepSeek R1 locally on your PC or Mac, you can do so easily with LM Studio and Ollama. Here is a step-by-step tutorial to get started.

Requirements to Run DeepSeek R1 Locally

To run DeepSeek R1 locally on a PC, Mac, or Linux system, your computer must have at least 8GB of RAM. With 8GB of memory, you can run the small DeepSeek R1 1.5B model fairly well, offering output at around 13 tokens per second. You can also run the 7B model, but it will consume around 4GB of memory, which might make your system slightly unresponsive.

With more memory, you could run 14B, 32B, and 70B models as well, but again, you need a faster CPU and GPU. Currently, most programs don’t utilize the NPU (Neural Processing Unit) to run local AI models. Instead, they mostly rely on the CPU, and in some cases (e.g. high-end Nvidia GPUs), the GPU is used for inferencing.

As for Android phones and iPhones, it’s recommended to have at least 6GB of memory to seamlessly run the DeepSeek R1 model locally. You can run DeepSeek R1 on your Snapdragon 8 Elite, or other 8-series and 7-series Snapdragon chipsets. Furthermore, you can find out how the Deepseek R1 and ChatGPT o1 models compare right here.

Run DeepSeek R1 on PC Using LM Studio

LM Studio is the easiest way to run the DeepSeek R1 model locally on PC, Mac, and Linux systems. It has a user-friendly interface and you can explore and download compatible AI models in a few clicks. On top of that, the application is completely free to use.

• Download and install LM Studio 0.3.8 or later (Free) on your PC, Mac, or Linux computer.
• Next, launch LM Studio and move to the search window in the left pane.
• Here, under Model Search, you will find the “DeepSeek R1 Distill (Qwen 7B)” model (Hugging Face).
• Now, click on Download. You must have at least 5GB of storage space and 8GB of RAM to use this model.

• Once the DeepSeek R1 model is downloaded, switch to the “Chat” window and load the model.

• Simply select the model and click the “Load Model” button. If you get an error, reduce “GPU offload” to 0 and continue.

• Now, you can chat with DeepSeek R1 on your computer locally. Enjoy!

Run DeepSeek R1 Locally on PC Using Ollama

• Go ahead and install Ollama (Free) on your Windows, macOS, or Linux computer.
• Now, launch the Terminal and run the below command to run DeepSeek R1 locally.
• This is a small 1.5B model distilled from DeepSeek R1 and based on Qwen for low-end computers. It only uses 1.1GB of memory.

ollama run deepseek-r1:1.5b

• If you have a large pool of memory with powerful hardware, you can run 7B, 14B, 32B, or 70B models distilled from DeepSeek R1. You can find the commands here.
• Here is the command to run the 7B DeepSeek R1 distilled model on your computer. It uses 4.7GB of memory.

ollama run deepseek-r1:7b

• Now, you can chat with DeepSeek R1 locally on your computer, right from the Terminal.
• To stop chatting with the AI model and exit, use the “Ctrl + D” shortcut.

Run DeepSeek R1 Locally Using Open WebUI

If you want to use DeepSeek R1 locally in a ChatGPT-like interface, you can install Open WebUI (GitHub) on your PC or Mac. It uses Ollama’s instance to offer several neat features such as Code Interpreter, voice chat, file analysis, custom instructions, and more.

Basically, you can have DeepSeek R1 running locally on your computer with all the same features as ChatGPT.

• First of all, go ahead and set up Python and Pip on your computer.
• Next, open Terminal or Command Prompt and run the below command to install Open WebUI. This step will take several minutes.

pip install open-webui

• Once installed, run the DeepSeek model via Ollama in the Terminal.
• You can follow the instructions mentioned in the above method to run Ollama.

ollama run deepseek-r1:1.5b

• After Ollama is up and running, run the below command to start the Open WebUI server.

open-webui serve

• Next, click on http://localhost:8080 to launch the local Open WebUI server.
• Click on “Get started” and set your name here.

• Now, you can use DeepSeek R1 in a user-friendly interface.
• Open WebUI will automatically pick the DeepSeek R1 model in the drop-down menu.

• If you want to close Ollama and Open WebUI, right-click on Ollama under the system tray and quit the application. You can now close the Terminal window.

Run DeepSeek R1 Locally on Android Phone and iPhone

You can quickly run DeepSeek R1 locally on your Android phone and iPhone. On Android, I tested several apps, including LM Playground, Private AI, Llamao, and others, but PocketPal offered the best way to run local AI models on Android phones – all for free.

The best part is that PocketPal AI is also available on iOS, and it doesn’t cost a dime, unlike Apollo AI and Private LLM. On that note, here is how to get started.

• First and foremost, install the PocketPal AI app.
• Next, launch the app and tap on “Go to Models.”
• Here, tap on the “+” button at the bottom right.
• Choose “Add from Hugging Face” here.

• Search “deepseek” and scroll down to find “DeepSeek-R1-Distill-Qwen-1.5B” by bartowski.
• Open it and download the suitable model based on your phone’s available memory. I have downloaded the “Q5_K_M” quantized model for this tutorial as it consumes around 1.3GB RAM.
• After the model is downloaded, go back and tap on Load.

• Now, you can locally chat with the DeepSeek R1 model on your Android phone or iPhone.

So, these are the different ways to install DeepSeek R1 on your computer and smartphone and chat with the AI model without an internet connection. In my brief testing, both the 1.5B and 7B models hallucinated a lot and got historical facts wrong.

That said, you can easily use these models for creative writing and mathematical reasoning. If you have powerful hardware, I recommend trying out the DeepSeek R1 32B model. It’s much better at coding and grounding answers with reasoning.