TypeSafe AI launches Jev, a fast, typed decision model for machines

TypeSafe AI, backed by $40 million, introduced Jev, a machine-native AI model that outputs typed probabilistic decisions rather than natural language. Built on a System One architecture employing Reinforcement Learning for Calibrated Decisions, Jev can return structured answers like choice probabilities or confidence scores within 70 ms to 500 ms, far quicker than typical LLMs. The company highlights cost efficiency, charging $0.042 per million input tokens and nothing for outputs, a fraction of the price of models such as OpenAI's GPT-5.6 Terra.

Demonstrations include the model playing Doom and routing a customer-service query with department probabilities. Co-founder and CEO Diogo Almeida, a former OpenAI researcher, positions Jev for real-time automation, AI-driven workflows, and large-scale classification tasks, emphasizing its reduced hallucination risk due to typed outputs.