Home / Technology / DeepMind's SIMA 2 Unlocks Next-Gen AI Capabilities
DeepMind's SIMA 2 Unlocks Next-Gen AI Capabilities
13 Nov
Summary
- SIMA 2 integrates Google's Gemini language model for advanced reasoning
- SIMA 2 can complete complex tasks in previously unseen environments
- SIMA 2 can self-improve based on its own experiences
On November 10th, 2025, DeepMind shared a research preview of SIMA 2, the next generation of its generalist AI agent. SIMA 2 represents a significant step forward, integrating the advanced language and reasoning abilities of Gemini, Google's large language model, to move beyond simply following instructions to understanding and interacting with its environment.
Like its predecessor SIMA 1, unveiled in March 2024, SIMA 2 was trained on hundreds of hours of video game data to learn how to play multiple 3D games like a human. However, SIMA 2 has doubled its predecessor's performance, uniting Gemini's capabilities with the embodied skills developed through training.
According to DeepMind researchers, SIMA 2 is a "more general agent" that can "complete complex tasks in previously unseen environments" and is a "self-improving agent" that can learn from its own experiences, a step towards more general-purpose robots and artificial general intelligence (AGI) systems.
The team has demonstrated SIMA 2's abilities in various virtual environments, including navigating photorealistic worlds and interacting with objects based on natural language instructions and even emojis. The agent's self-improvement capabilities allow it to learn new behaviors through trial and error, guided by AI-based feedback.
DeepMind sees SIMA 2 as a significant step towards unlocking more general-purpose robots, with the potential to tackle a wide range of intellectual tasks and learn new skills across different areas.
