MIRA releases a 4-player world model trained on Rocket League
MIRA is a built by General Intuition, Kyutai, and Epic Games. It predicts and continues game-like action for four players in a Rocket League-style setting. It was trained on 10,000 hours of .
The model has 5 billion . It can run a four-player scene at 20 frames per second on one B200 . A playable online demo, a detailed technical report, a 1,000-hour four-player gameplay dataset, and a are available.
An in-person demo at ICML lets people try it with PlayStation controllers.
Key points
- MIRA is a for four-player Rocket League-style gameplay.
- It was trained on 10,000 hours of .
- It has 5 billion and runs at 20 frames per second on one B200.
- The release includes a demo, technical report, dataset, and .
- It matters more for agent simulation than for token cost reduction.