Chinese AI model startup Naive AI, founded in February by a Tsinghua University professor, has been valued at more than $1.4 billion and raised $400 million from Tencent and others in three funding rounds, The Information reported on Sept. 18 local time.

Beijing-based Naive AI plans to release its first large language model as early as this month. It plans to make it available as an open-weight model like DeepSeek, Moonshot and Alibaba.

Naive AI is led by professor Dai Zhipeng (다이지펑), who is well known in China’s AI academia. He has studied deep learning, computer vision, foundation models and agentic AI over the past 10 years.

Naive AI is not building its LLM, called Naive, from scratch. Instead of costly pre-training, it is using an existing open-weight model from China as a base, but it has not been confirmed which model it is, The Information reported.

China’s LLM market has been led by big tech companies and DeepSeek, Moonshot and Z.ai. Against that backdrop, Dai believes there is also an opportunity for developers that are strong at boosting model performance in later stages such as mid-training and post-training, The Information reported.

Naive AI is focusing on raising performance across multiple tasks by changing the structure of a model that has completed pre-training and improving it through reinforcement learning and other methods. It is also conducting research into recursive self-improvement, in which an AI model improves its own performance, a core challenge for many AI developers including OpenAI, Anthropic and Google.

Developing an LLM by using another lab’s model is a tactic some U.S. AI companies also use. Thinking Machines Lab, led by former OpenAI executive Mira Murati, unveiled its first LLM, Inkling, in July and said it used the DeepSeek V3 open-weight model released in late 2024. Cursor, now owned by SpaceX, also developed its own coding model using a Z.ai model.