Jev is an AI model from TypeSafe designed to make fast, structured decisions rather than generate sequences of tokens.
Probabilistic decision-making itself is not new. Machine-learning systems have done this for decades. Recommendation engines, fraud detection systems and classifiers all make predictions based on probabilities.
The difference is that these systems are normally built or trained for a particular task.
Jev is intended to be more general. You give it some information, define the decision to be made, and it returns a structured answer with probabilities.
Modern LLM-based systems are general purpose. They can reason, calculate, use tools and take actions, but the underlying model still works by generating sequences of tokens.
Jev is an attempt to create a general-purpose model specifically for structured, probabilistic decisions.
It is also very fast. TypeSafe reports response times substantially faster than frontier LLMs on comparable decision tasks.
OpenAI and Google now offer Decisions APIs with similar uses. These return probabilities, choices from defined options or scores that software can use directly. Jev is a model designed specifically for these decisions; the APIs provide other ways to obtain similar outputs.
These capabilities can be combined. An LLM can handle language, open-ended reasoning and tool use, while Jev or a decision API handles fast decisions such as routing, scoring, checking or deciding what happens next.
That could be particularly useful in agentic systems.