The new model is up to twice as successful at some tasks.
DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"
In short
Google DeepMind has launched Gemini Robotics 2.5 to advance physical task execution. The newly introduced system demonstrates substantial performance improvements, achieving up to double the success rate of previous capabilities on certain tasks. Alongside the model, the lab made an open safety benchmark named ASIMOV-jailbreak available to evaluate the technology.
Key points
- The announcement pairs substantial robotic performance gains with an open safety testing…: The announcement pairs substantial robotic performance gains with an open safety testing benchmark.
- Gemini Robotics 2.5 doubles the success rate of prior capabilities on select operational…: Gemini Robotics 2.5 doubles the success rate of prior capabilities on select operational tasks.
- The release of ASIMOV-jailbreak provides an open standard to assess vulnerabilities in ro…: The release of ASIMOV-jailbreak provides an open standard to assess vulnerabilities in robotics models.