Jev is the first "System-One Model" from startup TypeSafe AI, founded by ex-OpenAI researcher Diogo Almeida (co-developer of the Instruction-Following methods behind ChatGPT). Unlike classical LLMs, Jev doesn't generate flowing text but instead delivers parallel-computed answers with calibrated probabilities for predefined, typed decision questions (Choice, Score, Noul). The model was trained using a novel method called RLCD (Reinforcement Learning for Calibrated Decisions) over a two-year stealth period. The model has been in early access via waitlist since September 15, 2026, accessible through a proprietary HTTP-API.

Sources (6)