January 27, 2025 - DeepSeek, a Chinese AI startup, has made headlines following the release of its new large language model, R1, on January 20. The model quickly rose to the top of the Apple App Store's Top Free Apps Chart, surprising industry leaders in Silicon Valley.
DeepSeek's R1 is designed for complex problem-solving and has been developed at a significantly lower cost compared to models from established companies like OpenAI and Meta. The company reported that it built and trained its V3 model for under $6 million using approximately 2,000 Nvidia H800 chips, which are less powerful than the H100 chips favored by U.S. firms.
Despite limitations due to U.S. sanctions preventing access to advanced chips, DeepSeek's performance is notable. Industry experts have acknowledged the startup's rapid advancements, with Marc Andreessen, a Silicon Valley venture capitalist, calling R1 a 'profound gift to the world' and likening it to a 'Sputnik moment' for AI.
Meta's Chief AI Scientist, Yann LeCun, emphasized the significance of open-source models, stating that DeepSeek's success underscores the advantages of open research. Meta recently announced plans to invest over $60 billion in AI development for 2025, aiming to maintain competitiveness in the evolving landscape.
DeepSeek's technology, while still considered behind that of OpenAI and Google, is gaining traction, with its models ranking in the top 10 on Chatbot Arena, a performance rating platform.