MIRA — World Model with Multiplayer
French startup kyutai has collaborated with Epic Games to create a neural version of Rocket League that supports multiplayer with 4 players. The model demonstrates high stability through simultaneous generation of multiple viewpoints.
MIRA — World Model with Multiplayer
French startup kyutai has collaborated with Epic Games to create a neural version of the game Rocket League. The model allows for a multiplayer session with 4 players, each of whom can be controlled either by the model itself or by a human.
The generation of multiple viewpoints simultaneously has significantly increased the stability of the model. The model can play against itself without artifacts for practically unlimited time (the creators ran the model for 24 hours with a context window of 4 seconds).
It was trained on 10,000 hours of bot gameplay, and a significant portion of the dataset has been made publicly available. The code for training the model is also accessible.
Why it matters
AnalysisThis technology opens new possibilities for multiplayer gaming, allowing players to interact with neural models. The open access to data and code fosters community development and innovation in the AI field.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
GPT-6 Astra reportedly edits another AI model's weights without retraining
GPT-6 Astra is claimed to modify only 0.016% of Qwen3.5-0.8B's weights without gradients or retraining, improving reasoning, math results, and instruction following.
GPT Image Generative Models Prompting Guide
OpenAI's gpt-image generative models are designed to produce production-quality visuals and fully controllable creative workflows. They are suitable for both professional design ta...
Claude Sonnet 4.6 vs GPT-5.5 vs Gemini 3.1 Pro: 2026 model comparison
Every few weeks, a new model comes out, and every time the same questions arise: should I switch? what changed what to pay for?
MIT researchers show GPT-6 Astra can improve other AI models by editing their weights
In an experiment, Astra raised Qwen3.5-0.8B's accuracy on MATH-500 from 7.9% to 48.1% by changing only 0.016% of the model's weights.