arXiv Notify me• Dec 4, 2025paper

SIMA 2: A Generalist Embodied Agent for Virtual Worlds

Executive Summary

Continue your research

Keep pulling the thread on Demis Hassabis.

0 quotes

The challenges of transferring skills learned in virtual environments to the complexities and physics of the real world remain significant.
The computational resources required to train and run such a sophisticated foundation model-based agent are likely substantial.

Development of highly capable AI assistants for gaming, simulation, and other virtual platforms.
Accelerating robotics research by using virtual worlds as scalable training grounds for physical agents.
Creating more natural and intuitive human-AI interaction through conversational, goal-oriented agents.

SIMA 2 is capable of reasoning about high-level goals, conversing with users, and handling complex instructions given through language and images.

In a diverse portfolio of games, SIMA 2's performance substantially closes the gap with human performance.

SIMA 2 demonstrates robust generalization to previously unseen environments.

SIMA 2 can autonomously learn new skills from scratch in a new environment by leveraging Gemini to generate tasks and provide rewards.

SIMA 2 is a generalist embodied agent designed to understand and act in a wide variety of 3D virtual worlds.

SIMA 2 is built upon a Gemini foundation model.

SIMA 2 retains the core reasoning capabilities of its Gemini base model.

Processed Jul 28, 2026Daily intelligence brief → yt-dlp + mlx-whisper + Gemini