Jev Cards

Jev Cards is a picture-quiz game where your opponent is a vision AI: can you beat TypeSafe's Jev model at recognising what's in a picture? Jev looks at a card and a question and picks an answer — or taps a place — in about a quarter of a second (median 261 ms on the first 200 cards), at roughly 70–80% accuracy. Your job is to be faster and more accurate than that.

There are four ways to play: Sprint (timed race), Versus Jev (head-to-head under time pressure), Streak (accuracy run), and Live match — start a room, invite friends, and Jev joins as one more player. Each mode has its own leaderboard. Cards are vector-drawn SVG with server-signed answers to stop cheating; the deck spans geography, chess puzzles ("white to move, mate in one"), fruit-spotting and more, with 300 new cards added days after launch.

The author, Jose Sabater (totisjosema on HN), is direct about the build. In his Show HN post (2026-10-20): "I made the cards through Opper's vision models and Claude Code." Asked in the comments whether it was vibecoded, he replied: "Thanks! Yeah built with claude code." The game is open source (the newest cards stay private for a while), free to play, 3 stars on GitHub, pushed 2026-10-21. Disclosure: the author works at Opper, whose site hosts this v2 build; Jev itself is TypeSafe's vision decision model.