What are people building with Jev?

Same test, same loop: Jev took 29 s and $0.003, Haiku 4.5 took 37 s and $0.07. Open source.

Jev reads every book index in parallel in 300 ms: "Who invented photography?" points to page 98, under Niépce.

Jev takes on the Pokémon League in Pokémon Gold (HGSS), one button press at a time.

Jev fights an open-model clone of itself in an FPS you can watch in the browser.

The sim hands Jev the scene as JSON and it picks the next bounded action (hover, descend, grasp, lift, place). "Red apple in the blue bin" takes 9 decisions.

Pulls market data, signals, risk and filings for any stock and has Jev call the next 30 days, with financial data from Valyu.

DeepSeek plans a sketch from primitives; Jev renders and refines it pixel by pixel.

120 people, 7 roads, 20 businesses, a council and a newspaper. Every hour Jev decides what each person does, works and spends.

A Chrome extension marks each post as clean, engagement bait, promo, secondhand or filler. $0.03 per 1,000 posts. Open source.

A 27K-parameter model parses queries into filters in 0.25 ms in your browser, and asks Jev only about the words it's unsure of.

Jev gets live data about the road and decides to change lanes, brake or accelerate around obstacles, pedestrians and red lights.

As soon as you stop typing, Jev rates the prompt; if it's simple, the app offers fast mode.

Maxfusion breaks a Meta video ad down; Jev judges each shot and rates the ad. 450+ ads in under 3 minutes.

Jev picks the agent, model, computer and folder for each task: Fable with Claude Code for a big rewrite, a Mac for iOS changes.

Jev powers the search in Replicas: results based on intent, and actions that adapt to what you're trying to do.

An open-source Snake arena where two Jevs compete; the low latency is what makes it playable.

As you write, Jev transmogrifies your sentences into something far more ornate.

Jev with the Cal.com API finds the best overlap between teammates instantly; the API round trips are the slowest part.

Jev picks every move, and every time Mario dies the whole VM forks into 4 timelines. Whichever Mario survives becomes canon.

Prompt classification in ~140 ms decides which LLM gets each request: 20x cheaper and 6x faster than before.

No more nested if/then logic: Jev picks the next question, json-render draws it and xstate keeps the state.

With a party of level-40 Pokémon, all for under $0.03.

Jev reads the prompt and picks the best of 50+ image and video models on Higgsfield.

Describe the form out loud and Jev builds it, fast.