Introduces a model type that answers from a predefined set with calibrated probabilities instead of free text. Engineers can use it for routing and classification without output parsing or hallucinated formats.
Key takeaways · AI-distilled
TypeSafe says Jev is calibrationHow well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.Full definition → so that a 0.8 answer is right about 80% of the time for that kind of answer, but only on average across many answers; any single answer can still be wrong.
Jev answers four question types, called primitives: Choice, Score, Noul and Box. A request has two parts: state, the text or images to evaluate, and a questions object with one or more typed questions.
Since Jev 1.14 the model accepts images as well as text, but per the post's comparison table it does not take audio or video input.
On OpenRouter, Jev costs $0.042 per million input tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → and output is free, according to the post.
The post frames Jev and LLMs as complementary: use an LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → when words are the output, such as a reply or code patch, and Jev for code-actionable decisions like routing and verification.
Terms in this piece · Glossary
calibration — How well a model's confidence matches reality — a calibrated model saying "90% sure" is right about 90% of the time.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
