China’s AI that Crushed OpenAI - How to Use DeepSeek R1 Locally

As someone who’s always curious about the latest advancements in AI technology, I discovered DeepSeek. It’s an AI model that has been making waves in the tech community for the past few days. In this article, I’ll share my experience with DeepSeek, covering its features, how it compares to ChatGPT and a practical guide on installing it locally.

What Is DeepSeek?

DeepSeek is an AI chatbot and language model developed by DeepSeek AI. It works like ChatGPT, meaning you can use it for answering questions, generating content, and even coding. The latest version, DeepSeek, is designed to be smarter and more efficient.

One of its biggest strengths is that it can run both online and locally. That means you don’t always need an internet connection to use it. DeepSeek is also gaining popularity among developers, especially those interested in privacy and AI models they can run on their own machines.

Another version, called DeepSeek R1, is specifically designed for coding tasks. If you’re a programmer, this could be a helpful tool for writing and debugging code.

Core Functionalities of DeepSeek AI:

• DeepSeek’s flagship model, DeepSeek-R1, is designed to generate human-like text, enabling context-aware dialogues suitable for applications such as chatbots and customer service platforms.

• The DeepSeek-R1 model incorporates “chain-of-thought” reasoning, allowing it to excel in complex tasks, particularly in mathematics and coding. This reasoning ability enables the model to perform step-by-step problem-solving without human supervision.

• DeepSeek has developed methods to train its models at a significantly lower cost compared to industry counterparts. For instance, the DeepSeek-R1 model was trained for under $6 million using just 2,000 less powerful chips, in contrast to the $100 million and tens of thousands of specialized chips required by U.S. counterparts.

• Unlike many proprietary models, DeepSeek is committed to open-source development, making its algorithms, models, and training details freely available for use and modification. This approach fosters collaborative innovation and allows for broader accessibility within the AI community.

• In response to hardware constraints, DeepSeek has focused on maximizing software-driven resource optimization, enabling the development of efficient AI models without reliance on advanced hardware.

• Beyond its primary language model, DeepSeek offers specialized models tailored for various applications, including:

  • DeepSeek-Coder: Designed for code autocompletion and assistance in software development.
  • DeepSeek-Math: Specialized in mathematical problem-solving and computations.
  • DeepSeek-VL (Vision-Language): A multimodal model capable of understanding and processing both text and visual information.

  
  

Through these core functionalities, DeepSeek AI aims to make advanced AI technologies more accessible and cost-effective, contributing to the broader application of AI in solving real-world challenges.

DeepSeek V3 Compare vs ChatGPT and Other Models

As you can see from the table below, DeepSeek-V3 is much faster than earlier models.

It is the best among open-source models and competes with the most powerful private models in the world.

Many people ask, “Is DeepSeek better than ChatGPT?” Here’s what I found:

1. Accuracy & Responses. DeepSeek V3 gives detailed answers, but sometimes, it feels less polished than ChatGPT. ChatGPT tends to be more refined in natural conversation, while DeepSeek is stronger in technical and multilingual tasks.

   
   

2. Local vs Cloud. One of the biggest advantages of DeepSeek is that you can run it locally. ChatGPT requires an internet connection, but DeepSeek V3 can work offline if you install it on your computer.

   
   

3. Features & Customization. DeepSeek AI models, especially DeepSeek R1, are great for coding. If you’re a developer, you may find DeepSeek R1 useful for writing scripts, debugging, and generating code snippets.

   
   

4. DeepSeek is a text model. It cannot produce images or videos.

How to Use DeepSeek V3

Option 1: Using DeepSeek V3 Online

• Visit the official DeepSeek AI website.
• Sign up or log in.
• Start chatting just like you would with ChatGPT.

If you want to turn on the DeepThink (R) model or allow AI to search when necessary, turn on these two buttons.

DeepSeek has a mobile app that you can also download from the website or by using this QR code.

Option 2: Running DeepSeek V3 Locally

If you want to run DeepSeek V3 on your own PC, follow these steps:

1. Visit https://ollama.com/, and download the Ollama app that lets you run LLM on your computer. It’s available for Mac, Windows, and Linux.

   
   

   

   
   

   I would recommend you use a terminal because it’s easier and faster. Just copy the command, and paste it inside the terminal window.

   
   

   

   curl -fsSL https://ollama.com/install.sh | sh
   

   
   

2. Then go to the Models page. The link is at the top left corner of the Ollama website. Find the deepseek-r1

It’s available in many different versions. If you don’t have a strong computer, I recommend downloading the 8b version.

Copy the command from the screen, and paste it into your terminal window.

ollama run deepseek-r1:8b

3. After the download is completed, you can start chatting with AI inside the terminal.

4. To launch it again, use the previous command ollama run deepseek-r1:8b

Keep in mind the offline model trained on data before pre-October, 2023.

Hardware Requirements for Running DeepSeek Models Locally

For smaller models (7B, 16B), a strong consumer GPU like the RTX 4090 is enough. Larger models (67B+) need multiple enterprise-grade GPUs like A100 or H100.

Video Tutorial

Watch on YouTube: 

Conclusion

If you’re curious about DeepSeek or looking for a new AI tool to add to your arsenal, I highly recommend giving it a try.

If you did, please share your feedback in the comments below.

Cheers! ;)