(Yicai) Jan. 27 -- DeepSeek-R1, a large language model recently released by Chinese artificial intelligence startup DeepSeek, has drawn the attention of industry professionals in Silicon Valley with its cost-effective and open-source characteristics, getting positive feedback and sparking debates on LLM competition.

Several scientists and investors from California's global innovation tech center have given DeepSeek-R1 high praise. "Open source models are surpassing proprietary ones," Yann LeCun, Turing Award winner and chief artificial intelligence scientist at Meta, said on X on Jan. 25.

Launched by Hangzhou-based DeepSeek on Jan. 20, DeepSeek-R1 topped the free app download charts on Apple's App Store in China and the US today, surpassing OpenAI's ChatGPT in the latter nation.

"DeepSeek has profited from open research and open source, for example PyTorch and Llama from Meta," LeCun noted. "They came up with new ideas and built them on top of other people's work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source," he said.

"DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets," noted Jim Fan, senior research scientist of US chip giant Nvidia. "They are perhaps the first open source software project that shows major, sustained growth of a reinforcement learning flywheel."

Marc Andreessen, the founder of Silicon Valley venture capital firm Andreessen Horowitz (a16z), described DeepSeek-R1 as "one of the most amazing and impressive breakthroughs I've ever seen -- and as open source, a profound gift to the world," on X on Jan. 24.

DeepSeek uses RL tech on a large scale in the post-training stage, improving DeepSeek-R1's reasoning ability with only little labeled data and making it comparable in performance to the industry-leading reasoning model OpenAI o1, according to the company.

In addition, DeepSeek-R1's application programming interface service is very reasonable, with only CNY1 (US 14 cents) per million tokens for input and CNY16 (USD2.20) per million tokens for output, much lower than OpenAI o1's USD15 and USD60, respectively.

Founded in 2023 by High-Flyer Quant, a Chinese quantitative investment fund, DeepSeek focuses on developing open-source AI models with low training costs. Last month, when DeepSeek released DeepSeek-V3, it said the training cost of the model was only about USD6 million.

Silicon Valley giants still pour huge amounts of money into the AI field. For example, OpenAI recently launched The Stargate Project, a program to invest USD500 billion in building AI infrastructure over the next four years.

Editors: Dou Shicong, Martin Kadiev