China Wind in Artificial Intelligence: What is DeepSeek?

MetaversePlanet February 13, 2025Last Updated: February 13, 2025

0 5 minutes read

The first month of 2025 has been very active for both the technology world and the cryptocurrency market. While OpenAI has long been the company most associated with artificial intelligence, it now faces serious competition from a Chinese firm: DeepSeek.

The AI industry has grown at a phenomenal pace over the past few years, driven by large language models (LLMs) and chatbots based on them. In particular, the ChatGPT chatbot released by OpenAI has revolutionized the field and reached hundreds of millions of users. Some studies indicate that ChatGPT processes more than 1 billion queries every day.

The OpenAI o1, which was demonstrated in September and fully released in December, stands out as the most powerful LLM to date. Although it operates slower than its predecessors, OpenAI o1 can solve much more complex problems and even exhibits the ability to “reason.” However, it now has a key competitor: the DeepSeek R1 model from the Chinese artificial intelligence company DeepSeek, which offers similar capabilities at much lower costs.

Thanks to its efficiency, DeepSeek-R1 has already established itself among the top AI projects. In this article, we will explore questions such as: What is DeepSeek? Is DeepSeek-R1 more powerful than ChatGPT and OpenAI o1? And is DeepSeek safe?

Contents show

What is DeepSeek?

DeepSeek is a China-based artificial intelligence company founded in 2023 by Liang Wenfeng. Wenfeng previously headed the AI-powered hedge fund High-Flyer. He then founded DeepSeek to develop open-source and accessible general-purpose artificial intelligence (AGI) applications. He took with him many scientists, engineers, and financiers who were experienced in this field.

Unlike most AI companies in China, DeepSeek operates independently of tech giants such as Baidu and Alibaba. The interesting thing is that, although it does not have the support of technology giants, it is even challenging the U.S. company OpenAI, let alone competing at the top of China’s artificial intelligence industry! DeepSeek’s newest LLM, the R1, works more efficiently than OpenAI’s o1, according to many third-party chip tests. While DeepSeek-R1 requires a much lower GPU to train than OpenAI o1, this increased efficiency also reduces the cost for the end user. Not just reducing it—but eliminating it entirely! DeepSeek-R1 is available for free, at least for now. That’s why it has the potential to become the biggest competitor to OpenAI and other U.S. AI applications.

What is DeepSeek-R1? It’s worth taking a closer look…

What is DeepSeek-R1?

DeepSeek launched its first LLM chatbot in 2023, shortly after it was founded. In May 2024, DeepSeek-V2 was released, and the company suddenly became the center of China’s artificial intelligence industry because it was much lower in cost than other companies’ LLMs. This sparked a price-cutting war among Chinese companies. Other tech companies such as ByteDance, Tencent, Baidu, and Alibaba began lowering the prices of their own AI models to compete with DeepSeek. DeepSeek, on the other hand, has managed to remain profitable despite its low price.

Six months later, in December 2024, DeepSeek-V3 was released. This model cost about $6 million to train. However, tests showed that it performed head-to-head with GPT-4o and Claude 3.5 Sonnet, which were created at much higher costs. DeepSeek-V3 even outperformed Llama 3.1 and Qwen 2.5.

However, the real sensation was DeepSeek-R1, which was released on January 20, 2025. DeepSeek-R1 was introduced as a competitor to OpenAI’s latest model, o1. What distinguished the o1 model from other LLMs was that it was trained with the reinforcement learning (RL) technique and had reasoning skills. A reasoning LLM can follow a train of thought before responding to the user, making logical inferences and arriving at very accurate conclusions.

DeepSeek-R1 stands out as an artificial intelligence model that is trained with the RL technique and can reason, just like o1. Users can access R1 for free after creating a DeepSeek account. R1 also shares in detail the reasoning process, allowing users to understand how it arrives at an answer before responding to their input.

DeepSeek-R1 vs. OpenAI o1

DeepSeek-R1 has made a huge impact in the world of technology and finance. So much so that the U.S. stock market opened today (January 27, 2025) with a sharp decline in the share values of technology giants, especially Nvidia shares. The success of DeepSeek in terms of price/performance and the fact that it can offer a service similar to OpenAI’s—which normally comes at a very high fee—for free has been interpreted as the transfer of competition between China and the U.S. into the artificial intelligence field.

Moreover, it suggests that DeepSeek-R1 has outperformed o1 in prestigious tests such as the American Invitational Mathematics Examination (AIME) and MATH. The performance of the two models in successfully solving complex problems can yield different results across different tests. However, it appears that both models perform close to each other. A study that analyzes leading tests with the support of artificial intelligence, the results of which you can see below, clearly reveals this.

Again, this study shows that DeepSeek-R1 and its variants promise much lower costs than o1 and its variants, even if their performance is equal. This provides a significant advantage for DeepSeek in terms of sustainability and efficiency. For the end user, DeepSeek makes a difference as a free alternative.

Benchmark Tests

Since its release, DeepSeek-R1 has been benchmarked against o1 in different disciplines. Let’s take a closer look at the results of these tests.

In math tasks, DeepSeek-R1 performs strongly. In the AIME 2024 test, which evaluates advanced multi-step mathematical reasoning, DeepSeek-R1 scored 79.8%, narrowly outpacing OpenAI o1-1217, which scored 79.2%. On the MATH-500, which tests LLMs with high school-level math problems, DeepSeek-R1 achieves an impressive score of 97.3%, while OpenAI o1-1217 remains at 96.4%.

There is also fierce competition between the two leading LLMs in coding tasks. In the Codeforces test, OpenAI o1-1217 comes out on top with 96.6%, while DeepSeek-R1 takes second place with only a small margin, scoring 96.3%. In the SWE-bench Verified test, which evaluates reasoning skills in software engineering tasks, DeepSeek-R1, which received 49.2%, outperforms OpenAI o1-1217, which received 48.9%.

When it comes to general knowledge, the two models are closely matched, but OpenAI holds the leadership position. GPQA Diamond, which measures the ability to answer general knowledge questions through factual reasoning, gives DeepSeek-R1 a score of 71.5%. OpenAI o1-1217, on the other hand, is the clear winner of this test with a score of 75.7%. In the MMLU, which measures multidisciplinary skills, OpenAI o1-1217 scores 91.8%, surpassing DeepSeek-R1, which scores 90.8%.

Is DeepSeek Safe?

DeepSeek quickly climbed to the top among AI companies and became OpenAI’s most important competitor. However, users and tech security experts have different opinions on DeepSeek’s data privacy and security policies. The answer to the question of whether DeepSeek is safe is actually directly related to the users’ approach to data security. Today, when subscribing to many AI tools, we allow them to use our inputs and data. Therefore, users who are concerned about their personal data should review DeepSeek’s privacy policy.

In this privacy policy, which you can access by clicking here, DeepSeek states:
“When you use our Services, we automatically collect certain information from you, including your IP address, unique device identifiers, and internet or other network activity information, such as cookies.”

It also collects user information entered when creating an account, such as date of birth, email address, username, and chatbot inputs.

If you don’t mind sharing such data, you can consider DeepSeek a safe AI tool.

What is Artificial Intelligence?

Artificial intelligence is a concept that refers to machines and software trained to mimic human intelligence. Any machine that exhibits characteristics associated with the human mind, such as learning, reasoning, and problem-solving, can be considered to have artificial intelligence.

There are also many artificial intelligence projects in the cryptocurrency market. AI projects that take advantage of blockchain technology, such as decentralization, security, and transparency, often issue and sell their own tokens to raise funding. These tokens, known as “AI coins,” which you can buy and sell on Bitlo, offer several benefits to their investors and make them part of an AI ecosystem.