Google DeepMind on Tuesday introduced Gemini Robotics 2.5, which it describes as its most capable vision-language-action model yet for giving robots an AI brain, alongside a separate next-generation model called AlphaGenome 2 that predicts how DNA sequences and their variations affect human cell types.
DeepMind said Gemini Robotics 2.5 improves dexterity, reliability and safety so engineers can move robots out of the lab and into dynamic, unstructured real-world environments where they work alongside people. The company said the model was tested across more than 200 distinct tasks and that it expects to scale testing further. It attributed part of the research acceleration to Gemini 4 Argon, its frontier large language model released in late September, which it said expanded the breadth and speed at which robots learn new skills.
The 2.5 release targets three areas. On dexterity, DeepMind said robots can perform more tasks with greater accuracy, including using everyday tools on delicate, flexible objects. On reliability, it said the model packed automotive parts with a 96% success rate during real-world proof-of-concept tests at Mercedes-Benz. On safety, it added improved physical steerability and a new benchmark for testing against jailbreaks. In evaluations, Robotics 2.5 achieved a 30% to 100% success rate depending on the task and completed tasks 20% to 30% faster than the previous version's baseline.
DeepMind said the model works natively with many types of robots, including Apptronik Inc.'s Apollo 2 humanoid robots and the Franka Duo platform with two-fingered parallel grippers. Apptronik co-founder and Chief Executive Jeff Cardenas said Apollo was built for useful work alongside people in real industrial settings, and that pairing Apollo 2 with Gemini Robotics produced dramatic model improvements within weeks of post-training that carried over to Mercedes-Benz production with little drop-off. Gemini Robotics 2.5 is rolling out through DeepMind's robotics partners, including Agile Robots SE, Apptronik and Boston Dynamics Inc.
AlphaGenome 2, also announced Tuesday, adds finer detail about specific cell types and states, which DeepMind said could reveal disease-related effects that broader tissue-level models miss. Such models help drug development and biotechnology companies turn genetic data into testable leads by predicting how DNA changes affect gene activity and which cells they act in. DeepMind said the predictions guide research rather than replace experimental validation, and that the model is not approved for clinical or diagnostic use.
DeepMind said it drew on the Human Cell Atlas, a scientific consortium that mapped gene expression in millions of cells, to show how different cells use genes, though predicting the variations that drive activation and expression changes remains a major challenge. Sarah Teichmann, a co-founder of the Human Cell Atlas and a professor of Stem Cell Medicine at the University of Cambridge, said the model's strength is adapting directly to new single-cell data without retraining, including from very small samples. DeepMind said AlphaGenome 2 matches the previous generation's predictions of gene expression, histone modification and chromatin accessibility while adding DNA methylation, a significant part of epigenetics.
More AI news from TechManNews.
