Jev, a decision-making AI model, outputs only numbers
Jev accepts text input but returns only floating-point numbers, letting it answer yes/no questions, choose from lists and score items.
A new type of model called Jev has been released, weeks after it first appeared, according to hackaday.com. Jev is designed for decision-making rather than conversation: it takes text as input but outputs only floating-point numbers.
That narrow output makes the model extremely fast and ultra-cheap to run, and it cannot write a single sentence. Instead, it classifies and categorizes quickly, handling user-specified yes/no questions, lists of choices and scoring-type requests.
Similar experiments have appeared since Jev's release, including Kev and Nimble. In October 2025, Nimble was added as a supported model in Ollama, the local model runner.
Because Jev returns a number rather than prose, its use cases sit closer to ranking and filtering than to text generation, though the facts provided do not specify benchmarks, pricing or an exact release date.
Quick answers
What is Jev?
Jev is a decision-making model type that accepts text input but outputs only floating-point numbers.
What can Jev be used for?
It answers user-specified yes/no questions, returns choices from lists and handles scoring-type requests.
Can Jev write text?
No. Jev cannot write a single sentence; it outputs numbers and is used for classification and categorization.