Jev

Jev is the first model from TypeSafe AI, released in early access on 15 September 2026, the day the San Francisco company came out of stealth with a $40 million seed round led by DCVC. TypeSafe calls it a "System One Model": it does not generate text. A developer passes in unstructured input, such as a support ticket or a document, together with typed questions whose possible answers are defined in advance, and Jev returns every answer at once, each with a calibrated probability and a confidence score, rather than writing a response token by token. TypeSafe says it therefore cannot make a type error, and that it was trained with a method it calls Reinforcement Learning for Calibrated Decisions. The intended uses are decisions inside ordinary software — classifying, routing, scoring, extracting, verifying and guardrailing — rather than chat. TypeSafe gives end-to-end response times of 70 to 500 milliseconds, and prices input at $0.042 per million tokens ($42 per billion) with output tokens free; the post itself notes the pricing cannot yet be shown not to be subsidised. Its "193.6x faster, 444.6x cheaper" comparison comes from TypeSafe's own workflow evaluations, which score agreement with the average answers of two other vendors' frontier models on workflows TypeSafe's team wrote, a set-up the post says may carry bias; these are vendor figures. In an independent hands-on test published by Every the same day, Jev read 37 documents and answered 21 questions about each in under 0.7 seconds; in a second test there it caught six of seven deliberately planted defects where Claude Fable 5.1 at high effort caught all seven, at a median 0.35 seconds per passage against 8.83. The founders are Diogo Almeida, a former OpenAI researcher, Erik Gafni and Sasha Sheng. Access is through a waitlist; no context limit, licence or weights release is stated.