Chaoyan AI Editorial Team

Recently, Chaoyan Intelligence Technology (Beijing) Co., Ltd. has successfully concluded its Angel and Angel+ funding rounds, amassing a total of nearly 400 million yuan.

The Angel round was co-led by IDG Capital, Starlink Capital, and Xingtai Technology, with additional contributions from Dexun Capital, Infinite Fund, Chuxin Capital, and Yunxiu Capital. The Angel+ round saw joint leadership from Zhongguancun Science City Fund, Shenzhen Innovation Investment Group, and Shanghai Future Industry Fund.

These funds will be allocated towards the development of foundational models and computing capabilities, research into recursive self-improvement technology, and the expansion of the team.

Founded in June 2026, the company announced the completion of these two funding rounds less than three months after its incorporation.

The lead investors in the Angel+ round are state-backed entities from Beijing, Shenzhen, and Shanghai, respectively. Xingtai Technology, which participated in the Angel round, is an AI-powered pharmaceutical firm listed on the Hong Kong Stock Exchange.

Chaoyan Intelligence is dedicated to self-evolving foundational large models, a concept known as Recursive Self-Improvement, where models autonomously enhance themselves, with a particular emphasis on scientific research and discovery.

The founder, Chen Yongchao, aged 27, serves as an assistant professor at the Tsinghua University School of AI. He earned his undergraduate and master's degrees from the University of Science and Technology of China and completed his Ph.D. through a joint program at Harvard and MIT. Prior to this, he conducted research at Google DeepMind and Microsoft.

Chen believes that models taught by humans are inherently constrained by human limitations.

His team comprises individuals from Tsinghua University, Peking University, Harvard, MIT, the University of Michigan, as well as industry giants like Google, Microsoft, Amazon, ByteDance, Kimi, and Zhipu AI.

From a technical standpoint, Chaoyan Intelligence initiates its approach at the foundational model level, guiding the model to generate its own ideas during both pre-training and post-training phases, thereby completing cycles of hypothesis, experimentation, validation, and iteration. Self-improvement is achieved during training iterations, rather than during real-time inference.

The inaugural product, Apex Research, has demonstrated remarkable capabilities. Out of 34 papers autonomously generated and submitted to the ACL Rolling Review, 11 received initial review scores above 3, with 2 outperforming 99% of human researchers. In the field of mathematics, the model has fully proven the dominance conjecture.

The company's long-term vision is for the model's scientific research capabilities to eventually surpass those of 100,000 top-tier AI scientists across diverse disciplines. The targeted application areas include chip design and simulation, molecular research and development, and quantitative strategy iteration, all of which are characterized by high value, high labor intensity, and high trial-and-error costs.

At present, self-evolution is confined to the training phase. Whether the model can consistently produce verifiable new results with minimal human intervention remains to be seen, and will require metrics beyond paper output for assessment. Questions also linger regarding how much computing power nearly 400 million yuan can support and when the first version of recursive training iteration will be finalized.