Chinese artificial intelligence (AI) startup DeepSeek is shaking up the world with its open-source AI model. It demonstrates that efficient AI development is possible even with low dependence on high-performance semiconductors, creating a new competitive landscape in the existing U.S. big tech-centric AI market.
According to IT media TechCrunch on the 26th (local time), DeepSeek noted that its inference-specialized model 'DeepSeek-R1', released on the 20th, surpassed OpenAI's 'o1' in performance tests. It also ranked first in the free app download chart on the Apple App Store, surpassing ChatGPT.
According to TechCrunch and the DeepSeek technical report, about 78 billion won ($5.576 million) was invested in the development of R1. This is one-tenth of the expense Meta used for its latest AI model development. DeepSeek trained the model using NVIDIA's low-spec chip H800 for export to China and achieved an accuracy of 79.8% in the U.S. Mathematical Olympiad benchmark test, surpassing OpenAI's o1 (79.2%). In coding tests, R1 recorded an accuracy of 65.9%, which is better than o1 (63.4%).
DeepSeek has made R1 and the 'DeepSeek-V3' launched last year open-source, allowing anyone to view, modify, and distribute the source code. The report also added that the API expense is priced lower compared to OpenAI, enhancing accessibility.
DeepSeek's achievements are shaking the existing AI competitive landscape, once characterized by 'the war of capital.' Satya Nadella, Chief Executive Officer (CEO) of Microsoft, said in an interview with CNBC, 'DeepSeek's model is impressive in terms of computing efficiency and expense reduction,' and added that 'we must take China's advancements seriously.'