Google's most capable AI agent for virtual 3D worlds
An Agent that Plays, Reasons, and Learns With You in Virtual 3D Worlds. Powered by Gemini, it goes beyond following basic instructions to think, understand, and take actions in interactive environments – meaning you can talk to it through text, voice, or even images.
SIMA 2 successfully operates in newly generated worlds created by Genie 3, demonstrating unprecedented adaptability. It can orient itself, understand instructions, and take meaningful actions in environments it has never encountered, advancing research toward general embodied AI.
SIMA 2 evolves beyond instruction-following to act as an interactive companion. It can answer user questions, reason about its own behavior and environment, and improve itself through interaction, making gameplay feel like collaboration rather than command-execution.
SIMA 2 demonstrates strong generalization by transferring concepts like 'mining' from one game to 'harvesting' in another, and successfully performing tasks in completely unseen games like MineDojo and ASKA. It closes a significant gap to human performance on diverse tasks.
SIMA 2 integrates Gemini's reasoning capabilities to understand high-level goals, reason about task steps, and execute goal-oriented actions in 3D game worlds. It can follow complex instructions, describe its intended actions, and adapt to new environments it has never seen before.