jevbooks

← All projects

PlayJev

OmniJev/PlayJev · Homepage

0.8B vision-language model plays ten browser games from raw pixels

Game & Simulation60%Runner-up: Agent Decisions

What it is

PlayJev is a fine-tune of Qwen3.5-0.8B-Base that plays ten classic browser games directly from raw pixels, aimed at real-time GUI-game agents and Jev-style decision models.

How it uses Jev

For each step the model receives the current frame, plus the previous frame when velocity matters, and the game's list of moves. Its single forward pass returns a probability over those moves; the game loop takes the argmax, and low-confidence steps can be handed to the per-game teacher.

Primitives:choice

Technique worth stealing

Shuffle the move list in every training sample so position carries no information.

Try it

python -m playjev.play snake --policy local --ckpt OmniJev/PlayJev-0.8B --episodes 1

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-centricyes0.84
  • Shows a System One patternyes0.85
  • Handles uncertaintyyes0.65
  • Measuredyes0.93
  • Runnableunclear0.53
  • Worth recommendingunclear0.45
  • Model replicayes0.96
  • Problem scopescore on a 0–2 scale0.05
  • About Jevyes0.70

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