Jev is the FIRST of a Whole New Class of AI Models

They come from the creator's own runs across a week, and the video is Firecrawl-sponsored and also promotes the creator's course. The page text documents no task definition, sample size, concurrency or percentile, so nothing published lets you reproduce the figure. Treat it as one practitioner's number, not an independent benchmark.

The copy says only that Jev played the creator's indie game at about three decisions per second. Nothing published documents the control surface, the decision set, whether the game loop was deterministic, or how many mistakes the model made. A game with a known finite action space is the friendliest possible test for a typed-decision model, so it does not transfer to open-ended UI work.

That is the thesis in the description, but the page records no before-and-after on a real repository. The closest evidence is a demo triaging every open pull request, and even that is described only as triage: no decision set, no false-positive count, no comparison against the regex or LLM call it supposedly replaced. Measure on your own labels before rearchitecting anything.

The chapters include a segment asking whether Jev is more than a classifier, but the page text does not document the answer. Nothing says how many dependent steps the demos chained, whether one question could build on another's answer, or how state was carried between calls. Typed questions are evaluated in isolation, which makes multi-step behaviour the part to scrutinise.

About three decisions per second. The copy records that figure for the game demo and separately reports roughly 200 ms per decision across a week of testing.

It is sponsored by Firecrawl and promotes the creator's course. The entry notes the decision numbers come from the creator's own runs, so read both the figures and the pitch with that in mind.

Speed, cost, reliability, community demos, and whether Jev is more than a classifier. The entry records the chapter list but not the answers each segment reaches.