PlayJev: a 0.8B visual decision model for ten games

An independent Jev-like model fine-tuned from Qwen3.5-0.8B-Base to choose legal actions from raw game frames. The project publishes model weights, training records and a gallery of ten browser games.

Open demo ↗

Notes

The author reports one forward pass per action, 43 ms per move on an H200, and training on 2.2 million frames; the model does not generate an action as text.

source-backed — Public repository, docs or live artifact. About this label

Destinations

Filed under

• #local-models
• #games
• #multimodal
• #open-source

In collections

• Games & simulations →
• Local & replica models →