2026-10-11 17:12 UTC

General Intuition, Kyutai, and Epic Games claim MIRA can interactively predict four-player Rocket League at 20 frames per second on one B200, providing an open testbed for multiplayer world-model prediction and control.

state: expiredheat: lowuncertainty: mediumknownscott: mediumworld-models physical-ai open-modelsGeneral IntuitionKyutaiEpic Games

What is this?

MIRA is a playable multiplayer world model created by General Intuition and Kyutai in collaboration with Epic Games. Trained on roughly 10,000 hours of bot-generated Rocket League gameplay, it predicts the dynamics and mutually consistent views of a four-player match in response to players’ inputs, reportedly running at 20 frames per second on a single Nvidia B200 GPU. An official GitHub repository releases code associated with the technical report, making MIRA a research testbed for interactive multiplayer prediction; however, the supplied snippets do not independently establish the report’s claim that it is the first such model.

Why it matters to Scott

The radar already tracks playable game world models on `radar:concept.world-models` and the closely related `radar:single-5090-playable-world-model`; MIRA extends that existing thread with mutually consistent four-player prediction rather than opening a new territory. Its released testbed bears on Scott’s hands-on game-agent experimentation in `dev:project.snake`, although the single-B200 requirement limits immediate local evaluation and the performance claims remain unverified.
dev:project.snakeradar:concept.world-modelsradar:concept.game-agentsradar:single-5090-playable-world-modelradar:concept.open-research
queries asked of Scott's wikis
  • interactive world models as agent environments
  • multiplayer simulation and multi-agent coordination
  • world models versus explicit game engines
  • open research artifacts and reproducible AI testbeds
  • real-time diffusion inference economics
  • synthetic gameplay data for embodied agents

Measured heat

no measured readings yet — the hourly heat pass fills this in

How the heat travelled

no chain yet — the hourly chain pass fills this in

Evidence (2) — ⭐ canonical anchor

sourceobjectauthorscorecomments
🟠 redditMIRA: Multiplayer Interactive World Models trained on Rocket League [R]
MachineLearning
MasterScrat10223
🟧 echo.paper ⭐The original primary artifact is the technical report, submitted to arXiv on July 6, 2026. It introduces “the first multiplayer world model Anthony Hu et al.; General Intuition, Kyutai, and Epic Games——

Interpretation history

Decision trace