Google DeepMind introduced Gemini Robotics 2.5 on November 10 and began rolling it out to robotics partners. The vision-language-action model is designed to translate perception and instructions into physical actions.

Task results need context

Google reports evaluation across 200 tasks, including manipulating tools and tying bows. Depending on the task, it claims 30%–100% higher success and 20%–30% faster execution than its prior baselines. These are vendor-reported comparisons, not independently established reliability for arbitrary environments.

A Mercedes-Benz automotive-kitting pilot using Apptronik Apollo 2 is a proof of concept. Its reported 96% success rate does not establish universal factory readiness.

Access is specific

Partner rollout does not establish open consumer access, a general API endpoint or published token prices. Robotics results also should not be substituted for language-model leaderboard scores.