TypeSafe AI Launches Jev, a Fast Decision-Making 'System One' Model

TypeSafe AI, founded by a ChatGPT co-inventor, launched Jev, a new category of AI model called 'System One'. Jev abandons text generation for structured, calibrated decision-making, achieving 70-500ms response times and significantly lower costs than traditional LLMs for classification and routing tasks.
Key insights
A New Model Category for Decision-Making
Jev is a 'System One' model designed for fast, structured decision-making rather than text generation. It outputs type-safe, calibrated results with confidence scores, making it suitable for classification, routing, scoring, and extraction tasks. Its architecture uses parallel sampling instead of autoregressive token generation, enabling 70-500ms response times, 20-200x faster than frontier LLMs, and 40-400x cheaper.
Trained with RLCD to Avoid Hallucination
Jev is trained using Reinforcement Learning for Calibrated Decisions (RLCD), a method developed by TypeSafe CEO Diogo Almeida, a co-inventor of ChatGPT. RLCD avoids the pitfalls of RLHF, which the founder argues incentivizes models to feign confidence. Because Jev does not generate free text, it cannot hallucinate. Every output is a structured decision with an epistemically honest probability.
Designed as an 'Intelligent If-Statement' for Developers
Jev is positioned as a component for developers to embed in system architecture, not a consumer product. It functions as an 'intelligent if statement' for tasks where rule engines are too brittle and LLMs are too slow or expensive. Use cases include smart branching in workflows, large-scale data processing, real-time applications, and acting as a guardrail or quality checker for other LLMs.