TypeSafe AI Unveils Jev, a System One Decision Model for Probabilistic Output

TypeSafe AI introduced Jev, a flagship "System One" AI model designed to take unstructured text state as input and return typed probabilistic decisions directly rather than generating text tokens. It evaluates questions in parallel and charges only $0.042 per million input tokens, making token output completely free. By eliminating auto-regressive text token generation, Jev achieves extremely fast processing times (70–500 ms) and significantly reduced costs for classification tasks like spam detection and search reranking. However, because it outputs only numbers without text reasoning, it pushes AI back toward black-box operations where hidden biases are harder to detect. Jev supports three typed question structures: Noul (Bernoulli yes/no confidence scores), Choice (categorical selection with probability distribution), and Score (numeric range evaluation). Users provide a single data state object along with multiple typed questions, which Jev evaluates in parallel within its context window.