jevbooks

← All projects

jev-playground

hegargarcia/jev-playground

Benchmarking Jev against other models in explicit-state games.

Evaluation & Benchmarking96%Runner-up: Game & Simulation

What it is

A Next.js playground for benchmarking TypeSafe AI's Jev against Luna, Haiku, Gemini, and Astra at choosing validated legal moves in tic-tac-toe and Connect Four, scoring decision quality and consistency across a sequence of moves.

How it uses Jev

For each model turn, the server supplies the board and every legal move as a Choice option with descriptive names, coordinates, and the resulting board. Jev uses a native evaluation model to choose among the given actions; the validated choice drives the next state.

Primitives:choice

Technique worth stealing

Give the model the full state plus every legal action as named Choice options, then validate the returned choice before applying it.

Try it

bun install, set AI_GATEWAY_KEY in .env.local, bun run dev, open localhost:3000/tic-tac-toe.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricno0.32
  • Shows a System One patternno0.31
  • Handles uncertaintyno0.03
  • Measuredno0.05
  • Runnableno0.08
  • Worth recommendingno0.29
  • Model replicano0.07
  • Problem scopescore on a 0–2 scale0.75
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.