Today, we’ll take a closer look at DeepSeek, a new language model that has stirred up quite the buzz. Let’s figure out what has everyone talking.
It seems the internet has a new favorite in the world of artificial intelligence, and it’s not the latest version of ChatGPT from the well-known OpenAI. Instead, this new contender comes from China, making waves across the AI landscape.
A Chinese AI chatbot has climbed to the top of the Apple App Store, becoming the most popular free application. This surge in popularity follows the release of the “thinking” model DeepSeek-R1 on January 20, which has surpassed OpenAI’s ChatGPT in downloads. Here’s a closer look at how this unfolded.
Read also: Can the U.S. Plant Its Flag on Mars? США
Why DeepSeek is interesting
This Chinese startup focuses on developing open AI models, allowing any developer to access the technology and build new models.
What sets DeepSeek apart from ChatGPT is its ability to articulate a chain of reasoning before providing an answer. User reviews on the Apple App Store and Google Play Store suggest that this level of transparency has been well-received by its audience.
The app is free to download and use, though users are required to register before gaining access to the AI.
On January 20, the Chinese startup DeepSeek launched its flagship AI model, R1, surprising Silicon Valley with the model’s advanced capabilities. R1 matched or even exceeded the functionality of AI systems developed by OpenAI, Google, and Meta—all while operating on a significantly smaller budget and without relying on the latest AI chips.
Over the past week, the app has gained significant traction among the public. It surpassed ChatGPT in popularity, reaching the #1 spot on Apple’s App Store in the U.S. and ranking among the top free apps for Android on Google Play Store at the time of writing.
The release of DeepSeek R1 has sparked questions about whether the billions of dollars spent on artificial intelligence in recent years were justified. It also challenges the perception that the U.S. remains the global leader in AI development.
Read also: Panama Canal: History of Its Construction and Basis of U.S. Claims
Who created DeepSeek R1
DeepSeek is an AI research lab based in Hangzhou, China, and R1 is its latest AI model. The company was founded in 2023 by Liang Wenfeng, an engineer and entrepreneur with a background in AI and finance. Before establishing DeepSeek, Liang led the private investment fund High-Flyer, which gained recognition for leveraging AI to analyze financial data.
The team at DeepSeek primarily consists of young graduates from top Chinese universities, including Tsinghua University and Peking University.
In March 2023, Liang’s fund announced via its official WeChat account that it was “starting over,” moving beyond trading to focus all resources on building a “new independent research group to explore the essence of AGI” (Artificial General Intelligence). Later that year, DeepSeek was established.
DeepSeek’s Large Language Model (LLM) first debuted in November 2023 as DeepSeek Coder, an open-source initiative. This was followed by the release of DeepSeek-V2 in May 2024. The company launched its latest model, DeepSeek-V3, in December 2024. Since then, the platform’s popularity has surged, with its mobile app surpassing 1.6 million downloads.
While DeepSeek LLM is largely similar to other popular chatbots, such as Google Gemini or ChatGPT, the app’s free models have gained significant popularity among users. Additionally, its developer-friendly API pricing has made it a focal point in discussions within the industry.
Read also: How Taiwan, China and the US are fighting for technological dominance: the great chip war
What is DeepSeek R1
DeepSeek currently offers two models: R1 and R1 Zero. At the moment, only R1 is available to users, though the differences between the two AI models are not immediately apparent. On paper, DeepSeek R1 is a general-purpose AI system, while DeepSeek R1 Zero utilizes Reinforcement Learning, meaning it is capable of fully self-training. However, these are technical aspects that may not be of much concern to typical users.
It’s important to note that DeepSeek R1 is an AI model developed by a Chinese company, and it stands on par with the latest available AI systems, such as OpenAI’s GPT and Anthropic’s Claude.
Read also: Nuclear Waste: What It Is and How It Is Disposed Of
DeepSeek R1 is free and unlimited
One of the key reasons DeepSeek has generated such a buzz is its cost for end users: it’s completely free. In a way, it’s the first highly advanced AI system available to users at no charge. For other systems like OpenAI’s ChatGPT and Anthropic’s Claude, a paid subscription is required, and even then, usage is often limited. Google’s Gemini is also available for free, but it’s restricted to older models and has usage limits. DeepSeek, on the other hand, appears to have no such constraints, making it fully accessible without restrictions for now.
Read also: 10 Mistakes That Simplify the Work of Hackers
DeepSeek R1 is extremely cost-effective
One of the reasons DeepSeek is free is likely due to what is said to be some key technological breakthroughs. One of the biggest advantages of DeepSeek R1 is its cost-effectiveness. While models like OpenAI’s ChatGPT o1 charge around $15 per million input tokens, the same tasks on DeepSeek can cost as little as $0.55. This low cost is probably why DeepSeek R1 is available for free to end users. It seems that DeepSeek has managed to optimize its AI system to such an extent that it doesn’t require massive computational resources or an abundance of graphics cards, keeping costs down.
In fact, efficiency is the key factor that has drawn attention to DeepSeek, especially in the U.S. Just a few weeks ago, such efficiency was considered impossible. But now, with DeepSeek demonstrating what can be achieved with just a few million dollars, AI companies like OpenAI and Google, which spend billions, are starting to look like real underachievers.
Read also: Transistors of the Future: New Era of Chips Awaits Us
DeepSeek R1 is open source
This is another key point that has attracted attention. Unlike OpenAI or Google systems, DeepSeek R1 is open source. Essentially, this means that DeepSeek not only makes major American tech companies look extravagant and irresponsible in their spending but also removes their ability to raise prices on their products. Ultimately, DeepSeek is essentially telling people that you don’t need to spend $1000 to access OpenAI or Anthropic systems. Instead, you can simply take this open-source model, customize it according to your needs, and use it however you want.
At the same time, the implications are not just business-related. They also have a geopolitical aspect. Instead of AI becoming yet another highly coveted and tightly guarded system owned by certain countries like the US, an open-source model like DeepSeek liberates technology that any country around the world can use to develop its own AI systems. It’s now clear that DeepSeek R1 is one of the most remarkable and impressive breakthroughs we’ve ever seen, and it’s a huge gift to the world.
Також цікаво: Neuralink Telepathy chip: what it is and how it works
How to get started with DeepSeek R1
You can use it in any browser by opening the link to DeepSeek R1, or download and install it from the Apple App Store or Google Play Store.
The most interesting part is that you can try DeepSeek R1 even without registering. However, it doesn’t work very efficiently in that case.
As for the smartphone app, users have recently been complaining that they are unable to register due to the high influx of people wanting to try the new Chinese model. I think this issue will be resolved soon.
Read also: All About Frontier Supercomputer
DeepSeek is increasingly effective in conquering the West
The sudden surge in popularity of the model is not coincidental. The dynamic development of artificial intelligence technology and the growing demand for advanced analytical tools have driven users to look for more accurate and efficient solutions. It turns out that DeepSeek has responded to these needs by offering a tool that not only processes data but also interprets its meaning within a specific context.
The key achievement of DeepSeek is the reduction in memory consumption and the acceleration of computations without significantly sacrificing the accuracy of results. The company has focused on streamlining the problem-solving process, avoiding detailed explanations of each step, which has significantly reduced computation time.
This allows the R1 model to demonstrate exceptional performance in mathematical and programming tasks, utilizing a chain-of-thought approach similar to that of ChatGPT o1.
Read also: Top Enterprise Meeting Software: Elevate Your Business Communication
DeepSeek vs. other language models: tests
It’s time to see whether the new model can actually pose a threat to the existing AI giants in the market. To assess this, we’ll consider external tests that have already been conducted.
However, it’s important to note that DeepSeek has not been included in many official tests, at least for two reasons. First, it is a relatively new player in the AI field. The R1 model was only introduced on January 20, 2025, meaning many earlier tests could not have included it.
Secondly, the Chinese company has applied a unique approach to training its model, focusing on software optimization and efficiency, which sets it apart from the traditional methods used by other models. This will likely lead to challenges when directly comparing it to other models in standard tests.
Hugging Face benchmark test
Let’s take a look at the tests conducted by Hugging Face, a company that develops computational tools for building machine learning applications.
DeepSeek-R1 achieves very high scores in many of the Hugging Face tests, outperforming models like Claude-3.5, GPT-4o, and even some variants of OpenAI o1 (though not all).
Key strengths of DeepSeek-R1:
- Thinking skills: In the English language category, it stands out for its impressive versatility and ability to provide accurate responses, particularly in MMLU and DROP tests.
- Coding and mathematics: In coding, the model shows exceptional performance, earning high scores on LiveCodeBench and Codeforces. In mathematics, DeepSeek-R1 outperforms competitors in tests like AIME 2024 and MATH-500.
- Multilingual support: Strong performance in both English and Chinese.
Benchmark artificial analysis from Artificial Analysis
Let’s look at further tests from Artificial Analysis, a company that provides independent analysis of AI models and API providers. This benchmark analysis examines the models from a slightly different perspective.
In the quality category, OpenAI o1 and DeepSeek R1 share the top spot in terms of quality, scoring 90 and 89 points, respectively, on the quality index. These models demonstrate the highest effectiveness in generating accurate and contextually relevant responses, making them leaders in this category. However, it’s worth noting that the difference between them, according to the test, is minimal.
In terms of token generation speed, DeepSeek R1 achieves 53.6 tokens per second, placing it ahead of OpenAI1 (32.3 tokens/second) and making it one of the fastest models in this category. However, it still lags behind models like ChatGPT o1-mini (210.5 tokens/second) and some versions of Gemini.
The next category is latency (time to first response). DeepSeek R1 has a latency of 29.88 seconds, placing it in the middle of the range. Models like Gemini 2.0 Flash (0.46 seconds) or GPT-4o (0.46 seconds) generate the first response much faster, which can be crucial for applications that require immediate feedback.
Read also: Most Fascinating Robotics Innovations of 2024
DeepSeek AI crashes markets
The entire market is in turmoil, and the main reason for this is the potential of the new technological revolution brought by DeepSeek AI, which clearly requires very low-cost infrastructure. Moreover, this new AI uses chips that are much cheaper compared to those used by American AI companies.
The efficient DeepSeek AI model has instilled fear among investors and traders worldwide, leading to a surge in the cryptocurrency and stock markets. Since the introduction of the AI, the prices of AI-based stocks and cryptocurrencies have risen sharply.
Companies working on AI algorithm development technologies have largely relied on expensive GPU chips. Enter DeepSeek AI, which uses affordable chips compared to other American AI companies.
This has sparked fear among investors who have invested significant funds in AI-based financial products. If the Chinese DeepSeek captures the AI sector, it could reduce the dominance of American AI companies in the market and lead to substantial losses for investors.
Major tech companies focused on artificial intelligence are already facing difficult times. Nvidia’s stock has dropped by more than 10%, dragging down other Western players like ASML. And this is just the beginning.
What’s next?
This issue is now troubling nearly the entire tech world. DeepSeek debuted as a blockbuster in the tech environment. Its emergence as a serious competitor in the AI space signals a shift in global technological competition. By using an economically efficient model and the open-source principle, it aims to disrupt the AI sector and dominate companies in the U.S. working in this field.
The success of the model has already been noticed in high political circles in China. On January 20, the day DeepSeek-R1 was unveiled, founder Liang Wenfen attended a closed symposium for businesspeople and experts organized by China’s Premier Li Qiang, according to the state news agency Xinhua.
Liang Wenfen’s presence at the meeting signals that the success of AI could be crucial to Beijing’s political goals of overcoming Washington’s export controls and achieving self-sufficiency in strategic sectors such as AI.
However, DeepSeek’s entry into the AI space has created tension in the industry, as the market fears its capabilities and highly efficient model. Experts believe that every new innovation brings its share of fears and changes. What will come of it remains to be seen. One thing is for sure, though—there will be no dull moments.
Read also:
- OpenAI Officially Launches Sora for Public Use
- Biomimicry: How Nature Inspires Engineers to Innovate