All news

25 September 2026

A week-old AI start-up pitching cheap, narrow models is drawing $10 billion investor offers

TypeSafe AI, founded by an ex-OpenAI researcher, launched Jev, a model that makes fast yes-or-no style decisions for software, and investors are reportedly offering to value it at $10bn or more.

Most of the AI race is about building ever-smarter chatbots. TypeSafe AI is betting on the opposite: a model that is not clever at conversation at all, just fast and cheap at making simple decisions. A week after its launch, investors are lining up.

What happened

TypeSafe, founded by former OpenAI researcher Diogo Almeida, came out of stealth (the quiet phase when a start-up builds without publicity) last week with a model called Jev. Its launch video on X drew 40 million views in under a week.

The start-up was valued at $200 million after a $40 million funding round more than a year ago. People familiar with the matter told the FT it is now getting offers that would value it at $10 billion or more. Almeida would not discuss new funding.

Jev does not write sentences or make images. Developers plug it into their software to make quick calls, such as whether to approve or block a request, where to send a support ticket, or how risky an insurance or credit applicant looks. Instead of reasoning at length like a chatbot, it picks from a short list of possible answers.

That makes it much cheaper. TypeSafe charges about 4.2 cents per million tokens (the chunks of text AI models process and bill by), against several dollars per million for large chatbot models, and says each query is roughly a hundred times cheaper and faster. Its lead investor, DCVC, says the company is already profitable.

Why it matters

Businesses are increasingly worried about what AI costs to run. If routine, high-volume tasks can be moved from expensive general-purpose models to cheap specialist ones, that could squeeze how much money leading labs like OpenAI and Anthropic make from those tasks. The name is a nod to Jevons paradox, the idea that making something cheaper can increase how much of it gets used overall.

Early signs of demand are strong. Hosting platform Vercel said Jev drew more than twice the interest from paid developer accounts in its first day of any previous model launch, and the AI gateway OpenRouter said requests to Jev more than tripled over the weekend.

There are skeptics. Arena CEO Anastasios Angelopoulos questioned how different Jev is from well-known classification tools, and Meta, Google and Hugging Face already offer similar ones. TypeSafe has kept details of how it trained Jev private.

What to watch

  • Whether TypeSafe closes a new round, and at what valuation.
  • How OpenAI and Anthropic respond on pricing for simple tasks.
  • Independent tests comparing Jev with existing classifiers.

Based on reporting by the Financial Times: The cheap new AI model taking aim at OpenAI and Anthropic. Facts, not advice.

More news