SIMA 2: A Generalist Embodied Agent for Virtual Worlds
Abstract
We introduce SIMA 2, a generalist embodied agent that understands and acts in a wide variety of 3D virtual worlds. Built upon a Gemini foundation model, SIMA 2 represents a significant step toward active, goal-directed interaction within an embodied environment. Unlike prior work (e.g., SIMA 1) limited to simple language commands, SIMA 2 acts as an interactive partner, capable of reasoning about high-level goals, conversing with the user, and handling complex instructions given through language and images. Across a diverse portfolio of games, SIMA 2 substantially closes the gap with human performance and demonstrates robust generalization to previously unseen environments, all while retaining the base model's core reasoning capabilities. Furthermore, we demonstrate a capacity for open-ended self-improvement: by leveraging Gemini to generate tasks and provide rewards, SIMA 2 can autonomously learn new skills from scratch in a new environment. This work validates a path toward creating versatile and continuously learning agents for both virtual and, eventually, physical worlds.
Keywords
Cite
@article{arxiv.2512.04797,
title = {SIMA 2: A Generalist Embodied Agent for Virtual Worlds},
author = {SIMA team and Adrian Bolton and Alexander Lerchner and Alexandra Cordell and Alexandre Moufarek and Andrew Bolt and Andrew Lampinen and Anna Mitenkova and Arne Olav Hallingstad and Bojan Vujatovic and Bonnie Li and Cong Lu and Daan Wierstra and Daniel P. Sawyer and Daniel Slater and David Reichert and Davide Vercelli and Demis Hassabis and Drew A. Hudson and Duncan Williams and Ed Hirst and Fabio Pardo and Felix Hill and Frederic Besse and Hannah Openshaw and Harris Chan and Hubert Soyer and Jane X. Wang and Jeff Clune and John Agapiou and John Reid and Joseph Marino and Junkyung Kim and Karol Gregor and Kaustubh Sridhar and Kay McKinney and Laura Kampis and Lei M. Zhang and Loic Matthey and Luyu Wang and Maria Abi Raad and Maria Loks-Thompson and Martin Engelcke and Matija Kecman and Matthew Jackson and Maxime Gazeau and Ollie Purkiss and Oscar Knagg and Peter Stys and Piermaria Mendolicchio and Raia Hadsell and Rosemary Ke and Ryan Faulkner and Sarah Chakera and Satinder Singh Baveja and Shane Legg and Sheleem Kashem and Tayfun Terzi and Thomas Keck and Tim Harley and Tim Scholtes and Tyson Roberts and Volodymyr Mnih and Yulan Liu and Zhengdong Wang and Zoubin Ghahramani},
journal= {arXiv preprint arXiv:2512.04797},
year = {2025}
}