DeepSeek: Key Points About the Chinese Startup, Comparison with ChatGPT

Due to its impact, the 500 wealthiest people in the world collectively lost $108 billion, the stock market experienced a significant drop known as "Black Monday," and technology companies began to worry about their future. This all stems from the launch of a new neural network called DeepSeek, which has dominated news headlines.

In this article, we explore what DeepSeek is, why it has generated such a buzz in the technology sector, and why its adoption is expected to become crucial. We also evaluate DeepSeek's performance in various marketing tasks and compare its results with those of ChatGPT to better understand its capabilities.

What is DeepSeek?#

DeepSeek-AI is a Chinese company specializing in the development of advanced AI technologies and large-scale language models (LLM). Its flagship product is the DeepSeek chatbot.

The startup was founded in late 2023 by Liang Wenfeng, the former head of the High-Flyer hedge fund. According to Western media, he used artificial intelligence to predict market trends.

Like ChatGPT or Gemini, DeepSeek is trained on textual data (books, articles, dialogues). Depending on the version, the model can be tailored to solve specific tasks, such as: generating a response for a user, solving mathematical problems, or writing sales copy.

Additionally, DeepSeek predicts the best response options by using patterns learned during training. The model takes into account the context of the conversation (previous replies), the semantics of the query (the hidden intentions of the user), and the accuracy of the information, eliminating errors with filters.

Recently, the company introduced two new models:

1. DeepSeek-V3 — introduced in December 2024. It uses the same algorithms as ChatGPT: generating text and answering questions. During tests, the model outperformed systems such as Meta's Llama 3.1, OpenAI's GPT-4o, and Anthropic's Claude Sonnet 3.5. What particularly surprised experts was the claim from the developers that the model took only two months and $5.5 million to train, whereas GPT-4 required over $100 million.

2. DeepSeek-R1 — introduced in January 2025. This advanced AI model is capable of reasoning, constructing logical chains, and independently verifying its conclusions, including intermediate ones. It can even pause to think through its response. In several tests, R1 showed better results than OpenAI's o1 neural network. According to the developers' statements, the model solves tasks in mathematics, programming, and logic as well as, and in some cases better than, its competitor. While the o1 version in ChatGPT is only available via subscription, DeepSeek-R1 offers users numerous free attempts.

DeepSeek also supports internet search (Search button) and can display code within the chat interface.

How DeepSeek Crashed the Stock Market#

DeepSeek became the cause of the "Black Monday" on the stock market. Bloomberg reports that on January 27, stock prices of tech companies dropped due to the success of the new neural network, and the 500 richest people in the world lost $108 billion.

The Nasdaq 100 Index dropped by 3% due to the success of the Chinese startup. This led to a loss of nearly $1 trillion in the market capitalization of companies within the index, according to Bloomberg.

Shares of Microsoft, Tesla, Nasdaq, and Broadcom fell, Nebius Group's (formerly Yandex N.V.) shares dropped by 37%, and NVIDIA, a leading supplier of chips for artificial intelligence, was particularly affected. The company lost nearly $500 billion in market capitalization. However, NVIDIA acknowledged DeepSeek's achievements and called the neural network "an excellent step forward in the field of artificial intelligence."

However, The Wall Street Journal calls the market drop a temporary situation. According to the journalists, one startup won't sink US tech giants.

Despite the situation in the market, ChatGPT remains calm. It responds that it views the new neural network without concern and does not believe its competitor will be able to replace it.

Why DeepSeek Became a Sensation#

The DeepSeek model is actively discussed due to its low cost. Many have become accustomed to thinking that training neural networks requires huge investments.

DeepSeek has debunked the myth that large investments are needed for neural networks. The company revealed that training its new models cost $5.6 million — significantly cheaper compared to the $100 million — $1 billion range mentioned by the CEO of Anthropic last year.

The new neural network has raised several important questions:

• If DeepSeek achieved such results after spending only a few million, why do other neural networks require billions?
• How objective are the current valuations of companies involved in AI technologies?
• Are expectations regarding the future of Western tech corporations inflated?

Despite this, analyst Stacy Rasgon from Bernstein considers the data about DeepSeek's low costs "extremely misleading".

How to Register on DeepSeek#

1. Go to the website or download the app. DeepSeek is currently available:

• On iOS
• On Android
• On the official website

2. If you're using the web version, click on the Start now button to begin:

3. After that, the site will redirect you to the registration page. Now, you need to enter your email and create a password. A code with access should be sent to your email. Once you enter the code, registration will be complete, and you can start using the service.

However, we were only able to register on the second day of trying. On January 27, DeepSeek temporarily suspended registration outside of China due to a surge in new users. The reason was a massive cyberattack on the service.

Our first attempt to use DeepSeek was also unsuccessful: the service tried to load a response for 3 minutes before showing an error.

Realizing that thousands of users were trying to test the Chinese marvel at the same time as us, we waited a bit, and it helped — soon, the neural network started working normally again.

DeepSeek vs ChatGPT: Comparison#

These are our remarks from our perception and experience using DeepSeek.

1. Origin and Accessibility:#

• DeepSeek is a Chinese AI model that has gained attention for its cost-effective nature and accessibility in regions where services like ChatGPT might be restricted. It works without VPNs and supports users in countries like Russia, where OpenAI's ChatGPT may face restrictions.
• ChatGPT, developed by OpenAI, is an American-based model that is widely known for its versatility and has seen consistent updates, with features such as internet integration, document processing, and more. However, its access can be limited in certain regions.

2. Performance and Use Cases:#

• DeepSeek offers strong performance in tasks requiring reasoning, like problem-solving and decision-making. However, its responses can sometimes be overly verbose and not as polished as ChatGPT’s, especially in customer service contexts.
• ChatGPT tends to generate more natural-sounding, concise, and clear responses, making it more suited for tasks like customer service, creative writing, and general queries. It also processes tasks more efficiently.

3. Cost:#

• DeepSeek has been highlighted for its low-cost model compared to other advanced AI systems. Training its models cost around $5.6 million, which is a fraction of the costs incurred by models like GPT-4.
• ChatGPT, particularly GPT-4, is a premium product with a significant investment in its development and maintenance. Users often need to pay for access to advanced versions, adding to the overall cost.

4. Regional Availability and Limitations:#

• DeepSeek is more readily available in regions where OpenAI's models may be restricted, and it is free to use in most cases. However, as with many newer models, there are still some stability and accuracy concerns.
• ChatGPT has widespread adoption globally but may be subject to restrictions and subscription fees in certain locations. Despite these, it is more stable and has a longer track record of updates and refinements.

5. Integration and Customization:#

• DeepSeek is still evolving and may require more manual fine-tuning and customization to meet specific user needs. For instance, tasks related to customer support may require extra effort to perfect its tone and response length.
• ChatGPT, on the other hand, benefits from a broader ecosystem of integrations, like API support, and has a more established community of users and developers creating various plugins and use-case-specific solutions.

Conclusion#

DeepSeek made a splash in the market and made the wealthiest people in the world notice its impact.

But for users, the new neural network is far from perfect: many requests are handled more effectively and without glitches by ChatGPT. However, it is important to remember that DeepSeek is just starting its journey, and its creators will refine all inaccuracies in the future.

DeepSeek may push the development of other neural networks, which will strive to improve in order to outdo the new competitor. This means the AI market will continue to evolve, and we will receive increasingly advanced tools that will make our work much easier and more efficient.