Illustration, AI-generated from the story text. Tap a panel to read the story through it.

Google DeepMind released Gemini Robotics 2.5, its , on Tuesday, rolling out to robotics partners Agile, Apptronik and Mercedes-Benz. Across more than 200 tasks, Google says success rates rose 30% to 100% over its previous baseline depending on the task type, and tasks finished 20% to 30% faster.

In a proof of concept, Apptronik's Apollo 2 running the model sorted and packed parts on Mercedes production lines in Kecskemét and Sindelfingen at a 96% success rate.

Reasoning-time raised safe completion by 33% around obstacles and 43% around nearby people. Google also released ASIMOV-Jailbreak, an open of attempts to talk a robot into unsafe actions, on which it says Gemini 4 Argon resists 99.5%, up from 82% for Gemini Robotics ER 2.

Ars Technica notes the model runs on cloud servers, where still limits real-time use, and that the demo video is sped up twice.
Google DeepMind's Gemini Robotics 2.5 lifts task success by 30% to 100% and hits 96% in Mercedes plants
Google DeepMind released Gemini Robotics 2.5, its vision-language-action model, on Tuesday, rolling out to robotics partners Agile, Apptronik and Mercedes-Benz. Across more than 200 tasks, Google says success rates rose 30% to 100% over its previous baseline depending on the task type, and tasks finished 20% to 30% faster. In a proof of concept, Apptronik's Apollo 2 humanoid running the model sorted and packed parts on Mercedes production lines in Kecskemét and Sindelfingen at a 96% success rate. Reasoning-time steering raised safe completion by 33% around obstacles and 43% around nearby people.
Google also released ASIMOV-Jailbreak, an open benchmark of attempts to talk a robot into unsafe actions, on which it says Gemini 4 Argon resists 99.5%, up from 82% for Gemini Robotics ER 2. Ars Technica notes the model runs on cloud servers, where latency still limits real-time use, and that the demo video is sped up twice. SourcesAB
Sources
- AGemini Robotics 2.5 accelerates real-world physical intelligence (Google DeepMind)deepmind.google
- BDeepMind reveals Gemini Robotics 2.5 and open safety benchmark ASIMOV-Jailbreak (Ars Technica, via newsletter summary)arstechnica.com
This story ran in the Afternoon Brief of November 10th 2026.