DeepSeek, a Chinese artificial intelligence company, has gained significant attention in the tech industry with the release of its DeepSeek-R1 language model. The model, developed by hedge fund manager Liang Wenfeng’s team, reportedly matches the performance of OpenAI’s leading model while being trained at a fraction of the cost – approximately $5.6 million using 2,048 Nvidia H800 chips, according to company statements.
Unlike its competitors, DeepSeek has made R1 fully open-source, allowing developers to modify and adapt the model for various applications. The model is available through a free website and mobile app, offering features such as web search integration. Tech leaders including Marc Andreessen and Meta’s Yann LeCun have praised DeepSeek’s approach, with LeCun emphasizing that this development demonstrates the potential of open-source models rather than Chinese technological superiority.
Industry observers note that DeepSeek’s success stems from its research-focused approach and efficient use of limited computing resources, a necessity due to U.S. restrictions on advanced chip exports to China. The company maintains offices in Hangzhou and Beijing, employing primarily Chinese university graduates. However, the model faces some limitations, including content restrictions related to Chinese regulations, and questions remain about its ability to compete long-term as U.S. companies invest heavily in new AI infrastructure.
Sources: VentureBeat, Financial Times, VentureBeat