What is Jev? TypeSafe AI's System One model
TypeSafe AI introduced Jev, a non-LLM decision-making model that provides reliable confidence scores for automated workflows, unlike traditional AI systems prone to inconsistency.
Jev is a decision-focused AI model from TypeSafe AI that does not generate text or images, instead producing typed answers with confidence levels for questions paired with contextual state. Unlike large language models, it avoids hallucinations and offers stable confidence metrics, addressing a key limitation in automating workflows such as customer support routing. The model is designed for high-volume, repetitive decisions like document sorting or ticket tagging, where speed and cost efficiency are critical.
Jev operates through three question types: Noul for yes/no responses, Score for rubric-based evaluations, and Choice for selecting from predefined options, all paired with confidence scores. Users input the state—such as a candidate’s CV—and Jev processes the data against provided criteria, returning structured outputs. For example, in a CV screening scenario, Jev assessed a candidate’s LLM experience, technical depth, and career progression, each with associated confidence levels.
The model launched in September 2026 after two years of development and is framed as a System One model, emphasizing speed and instinctive decision-making. It is named after William Jevons, referencing the Jevons Paradox, which suggests that efficiency in decision-making could lead to increased usage and improved outcomes at scale. Jev is now publicly accessible via API, allowing integration into automated systems or workflows.
TypeSafe AI positions Jev as a cost-effective alternative to LLMs for complex decision-making tasks, such as model routing, customer support triage, or large-scale data labeling. By automating high-confidence decisions and flagging lower-confidence cases for review, Jev aims to optimize workflows without sacrificing quality or incurring high computational costs.