Three rounds in seven months
Naive AI, a Beijing language-model startup founded in February, has raised about $400m across three rounds at a valuation of $1.42bn, The Information reported. The rounds were $100m, $180m and $120m, and the backers include Tencent, IDG Capital and HSG, the firm formerly known as Sequoia Capital China.
The valuation has nearly doubled inside five months. Crypto Briefing reported that the company was valued at about $800m in April. It has fewer than 100 employees and has not released a model.
The founder is Dai Jifeng, an associate professor at Tsinghua University who previously worked at Microsoft Research Asia and at SenseTime.
What it says it will ship
Naive AI plans to release its first large language model as early as this month, with open weights that anyone can download and adapt, according to the reports. It is not pretraining from scratch. The company is building on an existing Chinese open-weight system and putting its effort into mid-training, post-training and reinforcement learning — the stages after the expensive base run.
That choice is the interesting part of the company rather than the valuation. Pretraining a frontier base model is the single largest cost in the business, and a growing number of Chinese labs have concluded that the work worth paying for sits downstream of it. Whether $400m is a reasonable price for a team taking that route is a question the first release will start to answer.
An unresolved dispute
Dai was a technical adviser to another company, MiroMind, until January. The two sides disagree about what he took with him. Dai has alleged pressure to relocate; MiroMind has said its core technology and intellectual property remained entirely its own and that nothing was licensed to a third party. No resolution or lawsuit has been reported publicly since April, according to Implicator.ai’s account of the reporting.
The valuation gap
The raise lands in a market where Chinese model companies are valued on a different basis from American ones. Implicator.ai notes that Chinese models generate roughly a tenth of the revenue OpenAI and Anthropic do while carrying far higher multiples — DeepSeek at about 163 times annual recurring revenue against about 34 times for OpenAI. A company with no product and a $1.42bn price tag is the sharpest version of that pattern.
What to watch
The concrete thing to watch is the model itself. If it arrives this month under a permissive licence, Naive AI will have gone from incorporation to an open-weight release in eight months on someone else’s base model. If it slips, the valuation is resting on a founder’s record and a funding market rather than anything anyone can download.
