How to Run DeepSeek Locally on Windows: Complete Ollama Setup Guide (2026)

DeepSeek’s open reasoning models have become one of the most searched local-AI topics of 2026, and running them on your own machine means free, private, unlimited use with no API bills. If you have a capable GPU, here is exactly how to get DeepSeek running locally using Ollama, the easiest tool for the job.

Why Run DeepSeek Locally?

Running DeepSeek on your own hardware means your prompts never leave your machine, there is no usage cap or subscription, and you can pick a model size that fits your GPU’s memory exactly. The tradeoff is a bit of setup time and the upfront hardware cost, but on a modern GPU that pays off fast if you use AI daily.

What You’ll Need

• A Windows, macOS, or Linux PC with a discrete GPU (NVIDIA recommended; 8GB+ VRAM for smaller DeepSeek models, 16GB+ for the mid-size ones)
• At least 15-20GB of free disk space per model you plan to download
• Ollama, a free, open-source tool that handles downloading and running models with one command

Step 1: Install Ollama

Head to ollama.com/download and grab the installer for your OS. On Windows, double-click OllamaSetup.exe and click Install;

on macOS, drag Ollama into Applications; on Linux, run the one-line install script from the download page. The installer sets up Ollama as a background service, so once it’s done, Ollama is ready system-wide.

Step 2: Pull a DeepSeek Model

Open a terminal (Command Prompt, PowerShell, or Terminal) and run one of the following, depending on your GPU’s VRAM:

ollama pull deepseek-r1:7b     # 8GB+ VRAM
ollama pull deepseek-r1:14b    # 12-16GB+ VRAM
ollama pull deepseek-r1:32b    # 24GB+ VRAM (RTX 4090/5090 class)

Ollama downloads the model weights automatically, this can take a few minutes depending on your internet speed and the model size.

Step 3: Start Chatting

Once the pull finishes, run:

ollama run deepseek-r1:32b

This drops you into an interactive chat prompt right in your terminal. Type a question, and DeepSeek-R1 will show its reasoning step by step before answering, that “thinking” trace is what makes R1 different from a standard chat model. Type /bye to exit.

Picking the Right Model Size for Your GPU

DeepSeek-R1 ships in sizes from 1.5B up to 671B parameters. As a rule of thumb, the model’s file size needs to fit inside your GPU’s VRAM for full-speed responses. On an RTX 5090 with 32GB of VRAM, the 32B variant runs comfortably fast; the 70B variant will also load but leaves less headroom for long conversations. On an 8-12GB card, stick to the 7B or 8B tags. If a model is too big for your VRAM, Ollama will still run it by offloading to system RAM, but generation will be noticeably slower.

Frequently Asked Questions

Is running DeepSeek locally free?

Yes. Ollama and DeepSeek’s open-weight models are free to download and run; you only pay for the electricity and the GPU you already own.

Do I need an internet connection to use it?

Only for the initial download. Once a model is pulled, Ollama runs fully offline.

Can I run DeepSeek without a GPU?

Yes, on CPU alone with smaller tags like 1.5b or 7b, but responses will be much slower than on a GPU.

What is the difference between DeepSeek-R1 and DeepSeek-V3?

R1 is a reasoning-focused model that shows its chain of thought before answering, ideal for math, logic, and coding. V3 is a general-purpose chat model tuned for faster, more direct responses.

Is my data private when using Ollama?

Yes, everything runs locally on your machine. Nothing is sent to DeepSeek’s or Ollama’s servers unless you explicitly enable a cloud feature.

Run DeepSeek Locally: Performance and Model Size Tips

When you run DeepSeek locally on Windows with Ollama, the model size you choose determines both speed and hardware requirements. The 7B distilled version runs comfortably on 8GB of VRAM, while the 32B and 70B variants need a high-end GPU with 24GB or more. For most laptops and mid-range desktops, the 7B or 8B distilled models offer the best balance of speed and reasoning quality.

Once you run DeepSeek locally with Ollama, you can compare it against other open models from the official Ollama model library. If you’re setting up Ollama for the first time, check our Ollama on Windows 11 install guide for the base setup steps before pulling DeepSeek.

Quantized GGUF versions (Q4_K_M) cut memory usage significantly with only a small accuracy tradeoff, making it easier to run DeepSeek locally even on 8GB-16GB systems.