FULL STORY

TypeSafe's Jev: Launch, Real-World Tests, and a $40M Raise Under Fire

TypeSafe launched Jev, a non-generative decision model outputting probabilistic judgments. Real-world tests showed an open-source 27B model matching it, and its $40M raise and calibration claims drew skepticism.

2026-09-21 ~ 2026-09-21 · 3 episodes · 6 posts

Episode 1 · TypeSafe Unveils Jev, a Decision Model That Outputs Typed Probabilistic Judgments Without Text (2026-09-21, 2 posts)

TypeSafe has released Jev, a non-generative model that skips text output entirely: given a program state and typed questions, it returns typed answers with probability distributions in parallel. It aims to replace prompting chatbots and parsing prose-like replies.

Episode 2 · Jev tested: prompt tips and an open 27B model matches it (2026-09-21, 2 posts)

Testing TypeSafe's Jev yielded two prompt lessons (e.g., always add a 'none of the above' option), and a local open-source 27B model matched Jev's accuracy within 1.4 points with even better calibration on sentiment tasks.

Episode 3 · TypeSafe Raises $40M but Calibration Claims Face Scrutiny (2026-09-21, 2 posts)

TypeSafe AI emerged from stealth on September 15 with $40M led by DCVC at a reported $200M valuation, founded by ex-OpenAI researcher Diogo Almeida. Its model Jev's calibration and no-hallucination claims face scrutiny as no calibration data has been released.