SIMA 2
SIMA 2 is Google DeepMind's next-generation AI agent powered by Gemini that can understand, reason, and take intelligent actions in 3D virtual environments while interacting naturally with users through text, voice, or images.
https://goo.gle/SIMA-2?ref=producthunt
Product Information
Updated:Nov 18, 2025
What is SIMA 2
SIMA 2 (Scalable Instructable Multiworld Agent 2) is the latest milestone in Google DeepMind's research on creating general and helpful AI agents. Building upon its predecessor SIMA, which could follow basic instructions in virtual environments, SIMA 2 integrates advanced capabilities of Gemini models to evolve from a simple instruction-follower into an interactive gaming companion. It can navigate and solve problems across a wide range of 3D virtual worlds, including commercial games like No Man's Sky, Valheim, and Goat Simulator 3, while being able to understand user goals, perform complex reasoning, and improve itself over time.
Key Features of SIMA 2
SIMA 2 is Google DeepMind's advanced AI agent that integrates Gemini's language and reasoning capabilities to operate in 3D virtual environments. It goes beyond simple instruction-following to understand high-level goals, perform complex reasoning, and execute actions across different games and virtual worlds. The agent can communicate through text, voice, or images, learn from experience, and improve itself over time without human intervention. It demonstrates significant improvements in task completion rates compared to its predecessor and can effectively operate in entirely new environments, including AI-generated worlds created by Genie 3.
Gemini-Powered Reasoning: Integrates Gemini's language model capabilities to understand context, plan actions, and explain its decision-making process while interacting in virtual environments
Multimodal Interaction: Accepts instructions through multiple formats including text, voice, images, and even emojis, making it highly accessible and versatile
Self-Improvement Capability: Can learn and improve its performance through self-directed play and feedback, without requiring additional human demonstrations
Cross-Environment Generalization: Successfully operates across different games and virtual environments, including never-before-seen worlds, by transferring learned concepts and skills
Use Cases of SIMA 2
Game Testing and Development: Assists developers in testing games across different scenarios and environments, potentially reducing the time and resources needed for quality assurance
Robotics Training: Serves as a platform for developing and testing robot control algorithms in safe, virtual environments before deployment in the physical world
Virtual Assistant Development: Provides a foundation for creating more capable virtual assistants that can understand context and perform complex tasks in 3D environments
Research in Artificial General Intelligence: Serves as a testbed for developing and studying general-purpose AI systems that can adapt to new situations and environments
Pros
Significantly improved performance compared to SIMA 1, with better task completion rates
Ability to operate in completely new environments without prior training
Can learn and improve independently through self-directed play
Cons
Struggles with very long-horizon, complex tasks requiring extensive multi-step reasoning
Limited memory window for interactions
Challenges with precise low-level actions and robust visual understanding of complex 3D scenes
How to Use SIMA 2
Note: SIMA 2 is not publicly available: According to the sources, SIMA 2 is currently only available as a limited research preview to select academics and game developers. It is not available for public use or testing.
Basic Interaction Methods: When available, SIMA 2 can be interacted with through text, voice, or drawing sketches on the screen to give instructions to the AI agent.
Game Environment Setup: SIMA 2 works by observing the game screen and using virtual keyboard/mouse controls, without requiring access to game code or APIs. It can work across various supported games like No Man's Sky, Valheim, Goat Simulator 3, etc.
Giving Instructions: Users can provide natural language commands or high-level goals for SIMA 2 to accomplish. The agent will use Gemini AI to understand the intent and break it down into actionable steps.
Collaborative Interaction: Rather than just following commands, SIMA 2 can engage in back-and-forth dialogue to clarify goals, explain its reasoning, and describe what it plans to do next.
Multi-Language Support: Instructions can be given in different languages and even using emojis, which SIMA 2 can interpret and act upon appropriately.
Complex Task Execution: Users can assign complex multi-step tasks, and SIMA 2 will break them down and execute them while providing updates on its progress and reasoning.
Self-Improvement Mode: The agent can learn from its experiences and improve performance through self-directed play, though this appears to be an internal training mechanism rather than a user-facing feature.
SIMA 2 FAQs
SIMA 2 is Google DeepMind's next-generation AI agent that integrates Gemini language model capabilities to play, reason, and learn in 3D virtual worlds. It can follow complex instructions, engage in conversations with users, and improve itself through trial and error.
SIMA 2 Video
Popular Articles
GPT-5.1 Update: What’s New, How It Compares to ChatGPT 5 & How to personalize your own ChatGPT
Nov 13, 2025
Nano Banana 2 Release Date & Features: What to Expect from Google’s Next-Gen AI Image Tool
Nov 11, 2025
Microsoft MAI-Image-1 Release: What It Is, Why It Matters, and How to Use Microsoft’s New In-House AI Image Generator
Nov 6, 2025
Sora Invite Codes Free in December 2025 and How to Get and Start Creating
Nov 6, 2025
Latest AI Tools Similar to SIMA 2
JustAHuman
JustAHuman is a gaming platform that rewards players for completing challenges while helping game creators process 3D assets through AI.
Sketcho
Sketcho is an AI-powered design tool that transforms sketches and ideas into high-quality professional designs through an intuitive interface.
Rendair
Rendair is an all-in-one AI-powered architectural rendering platform that offers quick, high-quality visualizations through both AI tools and professional 3D artists for architects, designers, and real estate professionals.
Triorama AI
Triorama AI is an AI-powered 3D product configurator platform that enables eCommerce businesses to offer real-time product personalization and visualization capabilities to their customers.
Popular AI Tools Like SIMA 2
Hunyuan3D-2.1
Hunyuan3D-2.1 is Tencent's first production-ready 3D asset generation system that combines fully open-source framework with physically-based rendering (PBR) texture synthesis to create high-fidelity 3D models from images or text descriptions.
Modelify
Modelify is an AI-powered mobile app that transforms 2D images into high-quality 3D models seamlessly on iPhone and iPad devices.
Leonardo.ai
Leonardo.ai is a powerful AI image generator platform that enables users to create high-quality art, images, and videos through innovative tools like text-to-image generation, AI canvas editing, and 3D texture generation.
Android XR
Android XR is Google's new extended reality platform that integrates Gemini AI into wearable devices like smart glasses and headsets, enabling hands-free access to apps, real-time translations, navigation, and contextual assistance.