Detect AI-generated content with Google’s new watermark tool
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 can understand and execute tasks from multimodal prompts including user-drawn sketches, emoji commands, and instructions in different languages, enabling flexible and intuitive human-AI interaction beyond text-only interfaces.
SIMA 2 integrates Gemini's reasoning capabilities to interpret high-level goals, reason about tasks, and execute goal-oriented actions in 3D virtual worlds. It can understand complex and nuanced instructions, including abstract concepts and logical commands, enabling it to perform long sequences of actions toward user-defined objectives.
SIMA 2 represents a significant milestone in embodied AI research, combining Gemini's world-class reasoning engine with the ability to perceive, understand, and take action in complex 3D interactive environments. It demonstrates progress toward artificial general intelligence with implications for future robotics and AI-embodiment.
SIMA 2 demonstrates significant generalization improvements over SIMA 1, successfully transferring concepts like 'mining' across games and performing tasks in completely new environments (MineDojo, ASKA) it was never trained on. It can also operate in procedurally generated worlds created by Genie 3, showing unprecedented adaptability.
SIMA 2 moves beyond simple instruction-following to become an interactive gaming companion. It can describe its intended actions, explain the steps it's taking to accomplish goals, answer user questions, and reason about its environment and behavior—making interaction feel like collaboration rather than command-execution.