.

Jev is an AI model built by Diogo Almeida, a former OpenAI researcher. Most AI language models you’ve heard of, like ChatGPT, are built to write paragraphs, answer questions in detail, or generate code. Jev does something different: it’s built to make fast, simple decisions, like sorting things into categories or answering yes-or-no questions, rather than writing anything long.

What’s a “zero-shot classifier”?

This is the category of AI that Jev belongs to. Normally, if you want an AI to recognize something, like a cat in a photo, you have to train it on thousands of examples first. A zero-shot classifier skips that step. It can look at something it’s never specifically been trained on and still make a reasonable call about what it is, based on general knowledge it already has. So if you gave it a rough description of a cat, it could identify one on the first try, without ever having been shown a labeled photo of one before.

Why it’s cheaper and faster

The source material gives two numbers: Jev is about 444 times cheaper and 200 times faster than a full-scale model doing the same task. The reason is that Jev doesn’t carry the overhead a model needs for writing essays or generating code. It’s stripped down to produce short outputs only, things like “yes,” “no,” or a single score, and skips the extended reasoning that larger models like GPT-6 or Claude Fable go through before answering.

A decent comparison: asking a big model to do this kind of task is like asking someone to sit down and write a thoughtful essay. Asking Jev is like asking someone to flip a light switch. Both get something done, but one takes a lot more effort and time than the other.