Summary
Jev is a new type of AI model built to make decisions rather than generate text, and Matthew Berman calls it the most exciting AI he has seen in a long time. He compares the moment to the early days of ChatGPT, when people were still working out what a raw, powerful tool could do.
The video explains how Jev works. Users supply a structured question in JSON with clear criteria, and the model returns a probability, such as 73% true for whether a hot dog is a sandwich. Its strength is making tens or hundreds of thousands of decisions quickly and cheaply, with pricing of 4.2 cents per million input tokens and free output tokens. It is not meant for open-ended writing or coding, though it can assemble a web page from a library of UI components in under a second.
Berman then shows eight community-built use cases, including a detector that rated anthropic.com as 26% AI slop, Unclutter, an ad and cookie-banner blocker, and an email prioritizer that sorted 100 emails in under half a second. He also demonstrates Jev inside Zapier automations, such as deciding whether to accept calendar invites, and semantic find-in-page search.
📺 Source: Matthew Berman · Published September 24, 2026
🏷️ Format: Showcase
