01.AI, a Chinese startup led by artificial intelligence researcher and investor Kai-Fu Lee, has received a valuation of more than $1 billion less than a year after launching.

Bloomberg reported the milestone on Sunday. Lee told the publication that 01.AI achieved the unicorn valuation after raising funding from Alibaba Group Holding Ltd’s cloud unit and a number of other investors.

Lee is the chief executive of Sinovation Ventures, a major venture capital firm based in Beijing. He launched the firm in 2009 after a stint at Google LLC, where he founded and later led the search giant’s Chinese business unit. The executive previously established Microsoft Research China, one of Microsoft Corp.’s main international research hubs.

Lee began his career as a computer scientist at Carnegie Mellon University. During his academic career, he authored more than 60 papers and developed Sphinx, a groundbreaking speech recognition system.

Lee’s new startup started recruiting employees in March and officially began business operations three months later. It now reportedly has more than 100 staffers, more than half of whom specialize in large language models. TechCrunch reported that 01.AI’s workforce includes former AI researchers from Google, Alibaba and other major tech firms.

The company’s first product is an open-source neural network called Yi-34B. It’s a large language model with 34 billion parameters that can process text in English and Chinese. In a recent benchmark test carried out by Hugging Face Inc., YI-34B outperformed a number of competing open-source models with significantly higher parameter counts.

Parameters are settings that influence how an AI system processes input data. The more such settings there are in a neural network, the more tasks it can perform. But that increased sophistication comes at the price of hardware efficiency: The largest language models can run only on expensive data center graphics cards.

To address use cases that prioritize hardware efficiency, O1.AI is reportedly releasing a language model with 6 billion parameters. On the other end of the AI complexity scale, the company plans to build language models with 100 billion to 400 billion parameters. Lee told TechCrunch that 0A.1I will start rolling out those models in the next 18 months.

The company plans to generate revenue by providing proprietary AI models alongside its open-source projects. According to Bloomberg, some of those proprietary models will be optimized for specific industries and use cases. Lee said O1.AI will enable customers to integrate its neural networks into their applications.

01.AI’s elevation to unicorn status comes a few weeks after two competitors marked major fundraising milestones. Zhipu, a Chinese startup focused on developing large language models, announced in mid-October that it has raised $300 million from Alibaba, Tencent Holdings Ltd. and other investors. A few days earlier, rival language model developer Baichuan disclosed that it has secured a $300 million round of its own.

Image: Unsplash