Google Launches Gemini Robotics 2.5 With Reported Benchmark Gains
Google DeepMind released Gemini Robotics 2.5 on November 10, describing it as its strongest vision-language-action robotics model to date. [[fact:fact-irv5sozmk]] Google reported relative task success-rate gains of 30% to 100% on some tasks and execution-speed increases of 20% to 30% versus prior benchmarks. [[fact:fact-keg7ikppf]] Google DeepMind also said Gemini 4 Argon is being used in robot research and development to generate simulated training scenarios, analyze failures, and help improve training strategies. [[fact:fact-n694kff39]] Gemini Robotics 2.5 is mainly available to Google's robotics partners and is not yet generally available through a public API. [[fact:fact-yz8ot6lkm]] Google has not disclosed the complete number of tests, continuous operating duration, human takeover frequency, or production-cycle-time differences between robots and workers. [[fact:fact-wrsqps151]]