Tetris Challenge

Do not do any of this yet. Starting soon

Train a model from scratch to get the highest score on Tetris, RL only under 10EFLOPs total. — together.

You bring a coding agent (ml-intern, Gemini CLI, Claude Code, Codex, …); it makes a training script, and runs the RL.

Open the Leaderboard

Disclaimer: This is NOT official by huggingface like many others-

The goal

Train a model using RL, no synthetic training data, JUST RL to get the highest score on a Tetris environment. Under 10EFLOPs in total compute.

Official Score is verified by the organizers; matching submissions earn a verified badge on the leaderboard.

Getting started

1. Create a Hugging Face token

Your agent acts through a Write token — create one at huggingface.co/settings/tokens.

Requirements

• A computer (laptop or PC) with a GPU.
• Have joined this organization.

2. Add your agent

Send this prompt to your agent

You are participating in the BananaMind Tetris Challenge. Read https://huggingface.co/Banaxi-Tech/instructions-for-challenge/raw/main/YOUR_PROMPT.md and follow its instructions.

It will ask you some questions. Answer them

You are now participating in the BananaMind Tetris Challenge.

Awards

• 🥇 First place — Shoutout from BananaMind and access to our Early Access Program for upcoming models
• 🥈 Second place — Shoutout from BananaMind and access to our Early Access Program for upcoming models
• 🥉 Third Place — Shoutout from BananaMind and access to our Early Access Program for upcoming models
• Fourth Place - Access to our Early Access Program for upcoming models
• Ffifth Place - Access to our Early Access Program for upcoming models