Google DeepMind launches Gemini Robotics 2.5 and AlphaGenome 2 models
Also covered by 4 sources
What happened
Google DeepMind has introduced two new AI models: Gemini Robotics 2.5 and AlphaGenome 2. Gemini Robotics 2.5 is designed to improve robotic dexterity, reliability, and safety, and was demonstrated with humanoid robots sorting parts at Mercedes-Benz factories. AlphaGenome 2 is a DNA sequence model that predicts how genetic variants affect gene expression across hundreds of human cell types, supporting disease research. Both models are now available for research purposes, with AlphaGenome 2 accessible through an API for non-commercial use and in private preview for commercial research.
Key details
- Gemini Robotics 2.5 enables robots to perform over 200 real-world tasks, including automotive parts handling and tying bows.
- Apptronik's Apollo 2 humanoid robots achieved a 96% success rate in sorting parts using Gemini Robotics 2.5 at Mercedes-Benz production lines.
- AlphaGenome 2 was trained on more than 100 million single-cell profiles from 68 tissues.
- AlphaGenome 2 nearly doubles the detection of impactful genetic variants compared to its predecessor.
- An open-source safety benchmark, ASIMOV-jailbreak, was released alongside Gemini Robotics 2.5.
Why it matters
Manufacturers using robotics, such as automotive factories, can now automate more complex tasks with higher reliability and safety, potentially reducing labor costs and improving efficiency. Biomedical researchers gain a tool that can identify more disease-relevant genetic variants in specific cell types, accelerating the discovery of disease mechanisms and targets for drug development. Companies and institutions involved in robotics and genetics research now have access to advanced AI models that expand their capabilities for both industrial automation and genetic analysis.