Google DeepMind releases Gemini Robotics 2.5
Google DeepMind has released Gemini Robotics 2.5, a robotics model built with its frontier Gemini models, including one called Argon that generates simulated practice runs to help robots handle new scenes and objects.
Google DeepMind announced Gemini Robotics 2.5, the latest version of its robotics model. The company says the release was "accelerated by frontier intelligence models like Gemini 4 Argon."
The material available to us covers only the closing sections of the announcement. Most of it is about Argon's role in training. According to DeepMind, Argon can take a real-world camera scene, build a realistic digital twin from it, and generate "thousands of varied, simulated practice runs." When those runs are combined with real training demonstrations, the company says, robots generalize to new scenes and objects more quickly.
DeepMind also describes Argon as a debugging tool. It can look at video footage of failed tasks, suggest why they failed, and then test likely fixes both in simulation and on physical robots.
The company calls the release "an important milestone on the path toward solving physical AGI." It names three areas of progress: dexterity, real-world reliability, and safety. Agile, Apptronik and Mercedes-Benz are credited as partners on the release, though the excerpt doesn't say what each one contributed.
The excerpt includes no benchmark results, availability or access details, or technical specifications. It also doesn't explain what Gemini 4 Argon is beyond its role in this work, or how version 2.5 compares with earlier Gemini Robotics models.