Google DeepMind Launches Gemini-Powered SIMA 2 AI Agent for Advanced Multi-World Gaming Tasks
Google DeepMind introduces SIMA 2, a Gemini-powered AI agent capable of advanced reasoning, self-improvement, and navigating complex virtual 3D worlds, while Microsoft announces GPT-5 as default in Copilot Chat for faster and smarter AI assistance.
Google DeepMind has unveiled SIMA 2, an advanced AI agent powered by its latest Gemini models, designed to excel in multi-world gaming tasks with sophisticated reasoning and self-improvement capabilities. Building on the original SIMA, which learned basic instruction-following in virtual worlds, SIMA 2 represents a major leap towards artificial general intelligence (AGI) by embedding Gemini’s powerful reasoning engine at its core.
SIMA 2 can understand complex human-language instructions in diverse 3D game environments — including previously unseen games like ASKA and MineDojo — and skillfully execute multi-step goals. Its enhanced architecture enables it to explain its thought process, reason about high-level objectives, and apply learned concepts across different games, such as transferring "mining" skills from one title to "harvesting" in another. Importantly, SIMA 2 has self-improvement abilities that allow it to autonomously generate new tasks, evaluate its performance, and learn from experience without requiring human intervention. Demonstrations showed the agent navigating complex scenarios in games like No Man's Sky while interpreting nuanced instructions, a breakthrough for embodied AI intelligence.
Meanwhile, Microsoft has announced that its AI assistant, Copilot Chat, will adopt GPT-5 as its default model. This update introduces dynamic model routing to balance faster responses with improved complex reasoning, ensuring users receive timely and accurate assistance. The integration of GPT-5 enhances Copilot Chat’s ability to handle sophisticated workflows across Microsoft 365 apps, reinforcing Microsoft's leadership in practical AI deployment.
Together, these developments from Google DeepMind and Microsoft mark significant milestones in advancing AI agents capable of general reasoning, adaptability, and real-world problem-solving, proving transformative for future human-computer collaboration in gaming, productivity, and beyond.
