Games are amazing for training and evaluating AI. From simple neural networks that play Mario to multi-modal LLMs that can perform whole playthroughs of complex games.
However, these environments are nearly impossible to reproduce without pirating games, which limits their usefulness for training and benchmarking. That's why we're building OpenMonsters, a Pokémon-inspired MMO game powered by WorldQL. You can get the code on GitHub.
This initial release includes a foundation the game:
draft note from Jackson: #1 is free, it's just Dreamlab. 2-4 are super easy. Battles could be inspired by https://github.com/smogon/pokemon-showdown/blob/master/sim/SIMULATOR.md or just done cleanroom
The vision is to build OpenMonsters into a community-led training and evaluation environment. With various scenarios testing computer use, economic evaluation, coding challenges, and all sorts of useful skills within the context of a multiplayer game!
To celebrate our launch and to motivate quick development, we will be hosting an OpenMonsters model training tournament this November before Thanksgiving! More details to come. TODO: Specific details before publishing.
This project was inspired by the Meta OpenEnv community. Let's make amazing environments available to everyone!