01.AI has moved quickly from little-known startup to one of the most watched names in open source AI. Its Yi-34B model, released in November, outperformed Meta’s Llama 2 and rose near the top of prominent AI model rankings.
The Beijing company is taking a different path from firms that keep their most important AI systems tightly controlled. By releasing models that developers can download and build on, 01.AI is trying to turn technical momentum into a global ecosystem.
A fast rise for Yi-34B
Meta changed the open source AI landscape last July when it released Llama 2, a model similar to the one behind ChatGPT, for broad use. A few months later, 01.AI answered with Yi-34B, an open source model that quickly became a benchmark contender.
Within days of release, Yi-34B reached the top spot on a Hugging Face ranking that compares AI language models across standard tests for automated intelligence. A few months later, modified versions of the model were still scoring among the strongest options available to developers and companies on Hugging Face and other leaderboards.
01.AI has also expanded beyond text. On Monday, the company launched Yi-VL-34B, a multimodal AI model that can process images and discuss what they contain.
The result is a startup that has become hard to ignore in a field still led in many ways by OpenAI, Google, Meta, Microsoft, and other major technology companies.
The open source bet
01.AI’s strategy depends on giving developers access to its models, then benefiting from the community that grows around them. That contrasts with OpenAI, Google, and most other AI companies, which tightly control their technology.
The company’s immediate goal is not only to win benchmarks. Kai-Fu Lee, the founder and CEO of 01.AI, says the startup wants to build the next generation of AI applications on top of language models.
Lee argues that productivity software should not simply imitate older tools. As he put it, “The next-gen productivity tools shouldn't look like Office anymore—Word, Excel, PowerPoint—that’s the wrong way to go.”
Inside 01.AI, engineers are experimenting with AI-first apps for office productivity, creativity, and social media. None of those apps have launched, but the company is positioning them for a global market, in a similar way to how Chinese-backed TikTok and Temu became top apps with US consumers.
- Yi-34B has become the company’s breakthrough open source language model.
- Yi-VL-34B extends the effort into multimodal AI by handling images.
- AI-first apps are the business goal Lee wants to build around the models.
Kai-Fu Lee’s long AI arc
Lee describes Yi-34B as the result of a career spent trying to make computers understand people more naturally. “This has been the vision of my whole career,” he says. “It's been too long that we've had to learn computers’ language—we really need systems that can understand our language, which is speech and text.”
His background gives 01.AI unusual visibility. Lee did early artificial intelligence research before founding Microsoft’s Beijing lab and later leading Google’s Chinese business until 2009. In Chinese, 01.AI is known as 零一万物, Ling-Yi Wan-Wu in Chinese, meaning “zero-one, everything,” an allusion to the Tao Te Ching.
After emigrating from Taiwan to the United States and attending high school in Oak Ridge, Tennessee, Lee studied computer science at Columbia and Carnegie Mellon universities. His PhD work involved a speech recognition system that was cutting edge for the time.
He joined Apple as a research scientist in 1990, moved to Silicon Graphics in 1996, and returned to China in 1998 to help establish Microsoft Research Asia. In 2005, he became president of Google’s search business in China, then left in 2009 to start Sinovation Ventures.
Through Sinovation, Lee backed Chinese AI startups including Megvii, an image recognition firm, and WeRide, a company developing autonomous buses, taxis, and street sweepers. In 2018, he published AI Superpowers, arguing that Chinese AI labs and companies would soon rival those in the US because of talent, data, and users.
Why developers are paying attention
Western AI builders have already noticed Yi-34B. Jerermy Howard, an AI expert who recently founded Answer AI, said, “For many things, it’s the best model we have, even compared to 70-billion-parameter ones.”
Hugging Face CEO Clément Delangue also praised the model shortly after its release. “It’s a really good model that a lot of people are building on,” he said at a briefing in November.
Delange said open source language models are improving quickly and can outperform OpenAI’s GPT-4 on some specialized tasks. He also said many leading open source models have come from outside the US, giving companies such as 01.AI a chance to benefit when developers build new work around their releases.
That ecosystem effect is central to the open source AI race. If a model is strong enough, outside developers can adapt it, improve it, and use it in projects the original company did not create. In the case of Yi-34B, modified versions from developers in the West began appearing and exceeding its performance on the Hugging Face model leaderboard.
The wider US-China backdrop
01.AI’s rise is also tied to a larger shift in the technology rivalry between China and the US. Lee has often advocated collaboration between the two countries, but the environment has become more difficult.
In 2019, Sinovation Ventures shut down its Silicon Valley office, citing growing challenges in doing deals with US firms. In October of that year, the US government imposed sanctions on Megvii over government use of the company’s face recognition technology.
With Yi-34B, Lee is again trying to build bridges through open technology. The model is proficient in both Mandarin and English, and it has become a foundation that developers outside China are using and modifying.
01.AI is still early. The company was founded in June of last year, has raised $200 million from Alibaba and others, and is valued at over $1 billion, according to Pitchbook. Its apps have not launched yet. But its open source models have already made the company a serious participant in the next phase of generative AI.