Every few months, venture capital finds a new reason to throw a fortune at a company that is barely a year old. This week's reason: a model that refuses to talk.
TypeSafe AI, founded in 2024 by ex-OpenAI and ex-Meta researchers, closed an $870 million Series A on October 9 that values the company at $7.5 billion. Andreessen Horowitz led the round, with Sequoia and existing backer DCVC participating, and a16z partner Martin Casado is taking a board seat. The dizzying part: TypeSafe was worth $200 million just 24 days earlier, when its $40 million seed round closed on the same day its model Jev launched.
Jev is a transformer, but it is not a large language model. It does not generate text at all. Instead, it returns probabilities — what TypeSafe calls "calibrated decisions" — answering three question types the company labels Choice, Score, and Noul, a probability estimate. The training method is called Reinforcement Learning for Calibrated Decisions, or RLCD.
The pitch is ruthlessly practical: computers do not speak prose. They speak numbers. TypeSafe claims Jev runs significantly faster and uses far fewer tokens than LLMs, positioning it for automating tasks rather than generating content. "We have been super good at human language for four years, but it's not useful for automation because computers speak a different language," co-founder Diogo Almeida told TechCrunch last month.
TypeSafe claims a third of Fortune 500 companies are already using Jev — a claim worth treating as a claim, since the company has named no customers. Its lead investor put the figure at 25 percent the same day, which tells you something about how seriously anyone is auditing these numbers.
The category is already crowded. OpenAI released its Decisions API on October 6, running on GPT-6 Luna. Amazon shipped an open-source clone, Strands Decider 2B, on September 30. Cloudflare introduced Clef and Clef-flash through Workers AI. Stanford and Nvidia researchers found their open CLM-8B model up to nine times faster than Jev on some tasks, though less accurate at tool-calling.
TypeSafe has not released Jev's weights or its full training recipe. At $7.5 billion on the strength of a 24-day-old product, investors are not buying traction. They are buying a lane — and betting the post-text era is real.
