Gemini Robotics 2.5 accelerates real-world general dexterity
DeepMind has introduced Gemini Robotics 2.5, a model update that reportedly doubles the success rate for certain robotic tasks compared to previous versions. Alongside the release, the company launched ASIMOV-jailbreak, an open safety benchmark designed to test the model's resistance to adversarial prompts. These tools aim to improve both the efficiency and the secure deployment of autonomous systems in real-world environments.
Covered by 3 sources
- GGoogle DeepMind Blog↗3d ago
- TThe Decoder↗Maximilian Schreiner2d ago
- AArs Technica↗Ryan Whitwam2d ago