GoKawiilGoogle DeepMind upgrades Gemini Robotics to 2.5, adds open safety benchmark
Google DeepMind released Gemini Robotics 2.5, a vision-language-action model that controls robots, claiming 30-100% higher task success rates and 20-30% faster performance than its predecessor across over 200 test activities. The company also partnered with Apptronik and Mercedes-Benz to trial Apollo 2 humanoid robots kitting automotive parts in European factories, reporting a 96 percent success rate, and introduced an open-source safety benchmark called ASIMOV-jailbreak.
GoKawiil's interpretation of the reporting above, not reported fact.
The factory trial with Mercedes-Benz suggests DeepMind is moving robotics AI beyond lab demos toward real industrial deployment, which could accelerate adoption of humanoid robots in manufacturing. Releasing an open safety benchmark indicates the company anticipates scrutiny over robots operating near humans and wants to shape standards for evaluating risk before wider rollout.
- Gemini Robotics 2.5 shows 30-100% higher success rates and 20-30% faster execution than the prior version in over 200 tasks.
- Apptronik's Apollo 2 robot tied a bow and achieved 96% success kitting parts in Mercedes-Benz factories.
- DeepMind released an open-source safety benchmark, ASIMOV-jailbreak, ahead of wider real-world robot deployment.
Source: arstechnica.com, 2026-11-10
Published there as: “DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.