Google DeepMind introduces SIMA 2, an advanced AI agent capable of reasoning and learning within 3D settings, described as a "milestone in creating general and helpful AI agents." The agent utilizes the Gemini model to interpret instructions and autonomously improve through self-directed learning. Early tests show enhanced performance, reducing the gap between AI and human players. SIMA 2 is available as a limited research preview for select academics and game developers, signaling potential future applications in robotics.