Founded by Kai-Fu Lee, the former head of Google China, 01.AI has developed the Yi family of models, which rivals those of OpenAI, Anthropic, Google, and Meta.
However, access to these models and the API is restricted, and China’s ongoing trade tensions with the U.S. pose considerable challenges for the company’s future.
In the generative AI race, U.S. companies are leading the way, but China is making efforts to keep up. Recently, 01.AI has begun to stand out from the pack, employing a strategy similar to Meta’s but facing unique obstacles.
A lot of experience. As noted by the Turing Post newsletter, 01.AI’s founder Kai-Fu Lee began working in artificial intelligence in the 1980s while he was a student at Carnegie Mellon University. He has since held senior roles at Apple, Microsoft, and Google. Over time, he became Google’s top executive in China.
The birth of 01.AI. Lee’s latest venture marks a new chapter. In March 2023, concerned with China’s lag in the generative AI race, he launched 01.AI (known as Zero One Everything in China) and quickly raised $200 million in funding, with Alibaba among the investors.
Open, not proprietary. 01.AI’s primary model, called Yi, is based on Llama, Meta’s open-source model. This choice, rather than developing a proprietary model like those of OpenAI or Google, offers clear advantages, especially for attracting new developers and projects.
Yi as an alternative to GPT-4. The Yi family of models debuted at the end of 2023. Since then, 01.AI has released models like Yi-Large, designed to compete with GPT-4, along with smaller models such as Yi-Vision and Yi-Coder, which are less powerful than GPT-3.5. The latest large model, Yi-Lightning, is positioned to rival today’s most advanced models, according to LMSYS.
Wanzhi. Another key development from 01.AI is Wanzhi, a productivity assistant with capabilities similar to Office 365 Copilot. Wanzhi supports creating spreadsheets, documents, and presentations. It also interprets financial records, takes meeting notes, and summarizes lengthy texts in both Chinese and English.
Testing the AI model isn’t easy. 01.AI initially planned to make its models freely available for development. However, in August, the company blocked access to its API, citing “business adjustments” without further explanation. A Chinese phone number now appears to be required to test the service.
Revenue. 01.AI’s business model mirrors that of many Western AI companies, mainly generating income by charging for API access. According to The Japan Times, in May 2024, API access cost $2.50 per million input tokens (user requests) and $12 per million output tokens (chatbot responses). By comparison, OpenAI charges the same for input tokens and $10 for output tokens in GPT-4.
A complicated future. Despite its efforts, 01.AI is currently spending more than it earns—a typical scenario for AI startups—and faces a fragmented market. According to the Turing Post, 01.AI “cannot escape the realities of operating within China’s regulated tech environment,” where local control can impede growth. The ongoing trade tensions with the U.S., which restrict access to essential hardware and software, add another hurdle 01.AI must navigate.
Image | TechCrunch
See all comments on https://www.xatakaen.com
SEE 0 Comment