Wire flash
Ex-OpenAI researcher Diogo Almeida's TypeSafe raises $870M at $7.5B valuation
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
Diogo Almeida, a former OpenAI researcher who co-created RLHF (the training method that made ChatGPT conversational), has raised $870 million for his new company TypeSafe, which builds an AI model called Jev that never generates text. The company emerged from stealth on September 15 with a $40M seed at a $200M valuation, and 24 days later a16z priced its Series A at a $7.5 billion valuation—a 37x increase. Jev takes language input and returns only a decision from a list of options with a probability attached, designed for production systems rather than chat interfaces. TypeSafe claims each decision costs about $3 versus roughly $372 for a frontier LLM, enabling AI to be placed behind every branch in code. The company reports that 29.4% of the Fortune 500 have adopted the technology in three weeks with zero marketing.
Source report
Diogo Almeida, the researcher who spent four years at OpenAI and co-created RLHF—the training method that transformed raw models into conversational AI—has raised $870 million for a new company that builds AI designed never to speak.
The company, TypeSafe, has existed publicly for just three weeks.
A Staggering Valuation Jump
- September 15: Almeida emerged from stealth with a $40 million seed round at a $200 million valuation.
- Thursday: a16z priced his Series A at $7.5 billion.
That is a 37x increase in valuation in just 24 days.
The Core Bet: AI That Doesn't Write
Almeida's model, Jev, does not generate text. Instead, it accepts language input and returns a single decision from a predefined list of options, along with a probability score.
The logic is straightforward:
- A fraud check does not require an essay.
- Software cannot read prose, which is why almost no AI currently runs inside production systems. Instead, it sits in chat windows, interacting with humans.
Cost Efficiency at Scale
TypeSafe's internal calculations show:
- One decision on Jev: approximately $3
- Same check on a frontier LLM: approximately $372
At $3 per decision, companies stop rationing AI calls. The model can be placed behind every branch in code.
Early Adoption
Despite zero marketing, 29.4% of the Fortune 500 have already engaged with TypeSafe in its first three weeks.
The paradox: Almeida spent four years teaching machines to sound human. His $7.5 billion bet is that the real money lies in machines you never hear from at all.
Source
aakashguptaNeutral / independent