TypeSafe AI launches Jev, a model that skips text and returns structured decisions

US AI startup TypeSafe AI has officially launched Jev, a model it calls a "System One Model" built for fast, low-latency repeated decision-making. Instead of generating natural language or code, Jev directly returns structured decisions.

Jev's output includes three fixed primitives: Choice (picking one from up to 255 options), Score (rating within a set range), and Noul (returning a calibrated yes-or-no probability). It uses a technique called "Reinforcement Learning for Calibrated Decisions" to align confidence with actual accuracy.

The company claims Jev is 20 to 200 times faster than frontier large language models on structured tasks, with costs as low as 1/40 to 1/400. Input tokens cost $0.042 per million, and output tokens are essentially free.

TypeSafe AI was founded by Diogo Almeida, a former OpenAI researcher, and this week announced a $40 million seed round led by DCVC, ending its stealth mode.