Tsinghua professor’s stealth LLM startup Naive AI valued above $1.4bn before its first model ships

A Beijing startup called Naive AI has been valued at more than $1.4bn without releasing a model anyone can download. Founded in February by a Tsinghua University professor, the company has raised $400m across three funding rounds, with Tencent among the investors, and its first large language model could arrive as early as this month. The plan is to give that model away as an open-weight release, free to download and customise, in the mould of DeepSeek, Moonshot and Alibaba. A nine-month-old company, a billion-dollar price tag and a give-it-away strategy: China's LLM race has a new entrant, and the stealth label is doing a lot of work.

Watch Desk analysis

What happened

Naive AI is Beijing-based and was founded in February by a professor at Tsinghua University, though the report keeps the founder anonymous. The Information attributes the figures to a person with direct knowledge of the matter: $400m raised in three rounds, Tencent among the backers and a valuation above $1.4bn. The company has not announced the round.

The model is the more concrete promise. Naive AI is preparing to release its first large language model as early as this month, and intends to publish it as an open-weight release, similar to the downloadable, customisable offerings from DeepSeek, Moonshot and Alibaba. That places the newcomer directly in the most price-sensitive, most closely watched tier of China's AI market.

The numbers

• $400m raised
  Three funding rounds since February, per a person with direct knowledge of the matter.
• $1.4bn-plus valuation
  A unicorn price tag for a startup with no public model yet.
• Tencent on the ticket
  The only investor named so far; the full backer roster is unknown.
• First model due this month
  Planned as an open-weight release people can download and modify.

Why it matters

Open-weight is where China's model market gets decided. DeepSeek built its reputation on free downloads that people could run and adapt, and Moonshot and Alibaba compete in the same tier. A newcomer choosing open weights from day one is a statement about how challengers now enter: ship downloadable weights, win on price-performance, and let customisation build the community.

The money is the second signal. $400m before a single public release means investors are underwriting a team and a promise rather than a product. In a market already crowded with strong free models, that is either a bet that there is room for one more serious player, or evidence that open-weight ambition now attracts capital before the weights exist.

Our read

Treat the numbers as reported, not confirmed: one source, speaking anonymously, with no company statement behind them. That is ordinary for a story of this kind and no reason to dismiss it, but the proof will be a download link, not a valuation.

The genuinely interesting question is whether an unnamed team can differentiate in the most brutally competitive tier of the market. If the model lands this month, independent benchmarks and licence terms will settle the argument within weeks. Until then, the $1.4bn measures expectations, not capability.

What to watch

• Whether the first model ships this month and arrives open-weight as planned
• Confirmation of the round and the full investor list, ideally from Naive AI itself
• The founder stepping out of anonymity, which would be a story of its own
• Early benchmarks and licence terms once the weights are downloadable

Discussion spark: DeepSeek reset the open-weight market with one strong free release. Is Naive AI's $1.4bn pre-launch valuation a bet that history repeats, or proof that open-weight ambition now gets funded before the weights exist?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.