News

DeepMind Unveils Gemini Robotics 2.5 and ASIMOV Safety Benchmark

Google DeepMind has announced Gemini Robotics 2.5, the latest version of its robotics-focused model. According to the company, the model is up to twice as successful as its predecessor on some tasks, though the announcement excerpt does not specify which tasks or how performance was measured.

DeepMind also introduced ASIMOV-jailbreak, an open safety benchmark for assessing jailbreak risks in robotics models. Releasing the benchmark alongside the model gives researchers a standardized way to evaluate whether systems can resist instructions that cause unsafe behavior.

The announcement reflects an effort to advance robotic capabilities while making model safety testing more accessible to outside researchers.

Sources