TypeSafe AI Launches Jev, a Structured Decision Model Without Text Generation

TypeSafe AI has released Jev, its first 'System One model' designed to output type-safe structured numerical data rather than text, specifically for classification, routing, and information extraction. The model features end-to-end response times between 70 and 500 milliseconds and is priced at $0.042 per million input tokens, with free output. The model utilizes a training method called RLCD (Reinforcement Learning for Calibrated Decisions). Its limitations include an inability to process images and a field cardinality limit of 255.