DeepSeek-R1: China's AI model challenges OpenAI
New AI model from China causes turmoil
“Deepseek-R1 is now live and open source and competes with OpenAI's Model o1.” With this challenge, DeepSeek is entering into competition with established AI models such as OpenAI and represents serious competition with the previously established AI models of large US companies.
With the release of its R1 model, the Chinese AI start-up has surprised the technology world and shaken the stock markets. This open-source model offers impressive performance at a low cost. According to current evaluation standards, DeepSeek offers similar services to the current models from tech giants such as Google and OpenAI and is even superior to them in some respects.
What is Deep Seek?
DeepSeek is a Chinese AI start-up founded in 2023 from Hedge fund manager Liang Wenfeng was founded. The company specializes in the development of advanced language models and artificial intelligence. DeepSeek uses open source models where the source code is publicly available and can be used by others for their own applications.
On January 20, 2025, the Chinese company DeepSeek presented its AI-powered chatbot R1. Within a very short time, the DeepSeek app reached number 1 of the most-loaded iPhone apps in the USA, overtaking ChatGPT. DeepSeek is also one of the most frequently downloaded apps in the German app store. Users can chat with AI, upload files and use them for web research.
What can Deep Seek do? Performance and cost efficiency
DeepSeeks R1 is one of the modern “reasoning models.” This means that it is able to describe its “thinking” and thus ensures better comprehensibility. The AI answers in loops, i.e. rethinks its own answer. OpenAI is also following this approach with its model presented in December 2024. Another function of the model is multimodal data integration, i.e. the processing of text, image, audio and video simultaneously. Deepseek's latest version of the Janus image generator should also be able to take on Dall-E 3 from OpenAI or Midjourney.
The Skills in Areas such as Mathematical Problem Solving, Programming and Natural Language Processing should be particularly impressive. In math benchmarks, DeepSeek Narrowly surpasses OpenAI in the accuracy of the answers. Programming skills are also at an expert level, DeepSeek surpasses 96.3% of human participants in tests.
It is remarkable that the new model should not only be more powerful than its western counterparts, but also significantly more cost-effective. While US Technology Giants such as Meta, Alphabet or OpenAI have invested hundreds of millions to billions of dollars to develop their models, DeepSeek estimates the cost of training AI at just around 5.6 million dollars. It is not known how high the total development costs really are. Since DeepSeek is part of the Chinese hedge fund “High” flyer, it can be assumed that the company had access to significantly more than the stated approximately 6 million dollars. According to recent research from SemiAnalysis, the actual cost of developing DeepSeek-R1 could be around 1.3 billion US dollars, taking into account investments in infrastructure and GPUs. One reason for the low development costs can be attributed to the efficiency of using Nvidia chips. According to their own statements, DeepSeek only needed around 2,000 such chips. These chips were also said to have been outdated, as they were apparently purchased before the export ban, which was enacted in October 2023. Competitors have sometimes used 16,000 or more chips to train their models. In response to this information, Nvidia shares experienced a dramatic decline. This shows the impact Deepseek's latest model has on the industry and competition.
Censorship mechanisms and privacy concerns
A major criticism of DeepSeek is the integration of censorship mechanisms, which particularly concern politically sensitive issues. For example, the model refuses to answer questions about the protests and massacre in Tiananmen Square in 1989, the persecution of the Uyghurs, or human rights in China. There are also concerns about data protection. It is assumed that user data is stored on servers in China. This raises questions about data security and privacy protection. According to media reports, German data protectionists and authorities want to advise on this and review Chinese AI. One option to make the use of the model more secure could be access to Microsoft Azure via an interface, for example. Shortly after the release of Deepseek R1, Microsoft made this model available on its Azure cloud computing platform.
China's AI Initiative
The success of DeepSeek represents only a small part of China's far-reaching goals in the area of artificial intelligence. According to “Beijing Artificial Intelligence Industry White Paper 2024" China has a dynamic and growing AI ecosystem, with capital Beijing far ahead. With over 2,400 companies and 46 publicly traded companies, which together reach a market capitalization of 4.3 trillion yuan (around 590 billion dollars), Beijing is a leader in both innovation and financial value creation. In 2024 alone, private investments in generative AI (GenAI) rose fivefold to 3.15 billion dollars. This was driven by government support and a strong focus on industry-specific applications.
Is DeepSeek the new OpenAI?
DeepSeek has taken a significant step in AI development with its R1 model and offers a cost-effective and powerful alternative to established systems. However, censorship mechanisms and privacy concerns raise questions that should not go unaddressed. It remains to be seen how these factors will impact the adoption and global deployment of DeepSeeks AI technologies. Some tech experts aren't sure how DeepSeek's deep capabilities are really better than those of market leader OpenAI. We can be curious to see what reactions the established AI Giants will follow and whether the current hype is justified. The World of AI remains extremely progressive and is changing rapidly. Companies must take a close look at which technologies they want to rely on.
