AI3 min read0 comments

DeepMind ships Gemini Robotics 2.5 to robot partners and AlphaGenome 2 for DNA work

At Mercedes-Benz, robots running the robotics model packed car parts with a 96% success rate in proof-of-concept tests

AI and Altcoins
Published
GOOGLE DEEPMIND · ROBOTICS: 96%

Key takeaways

  • Google DeepMind released Gemini Robotics 2.5 on 11 November through partners including Agile Robots, Apptronik and Boston Dynamics.
  • Robots using the model packed automotive parts at Mercedes-Benz with a 96% success rate in proof-of-concept tests.
  • DeepMind reports task success of 30% to 100% and completion 20% to 30% faster than the prior baseline.
  • AlphaGenome 2 adds single-cell detail and DNA methylation prediction, and is not approved for clinical or diagnostic use.
  • Apptronik's Jeff Cardenas said Apollo 2 improved markedly within weeks of post-training.

Robotics model goes to partners

Google DeepMind released Gemini Robotics 2.5 on 11 November, a vision-language-action model the lab describes as its most capable so far for giving robots an "AI brain". The model is being distributed through the company's robotics partners, among them Agile Robots, Apptronik and Boston Dynamics.

DeepMind says the update targets dexterity, reliability and safety, the three areas that decide whether robots can leave controlled labs for unstructured spaces where they work near people. The lab tested the model on more than 200 tasks and said it expects to widen that testing further.

In proof-of-concept work at Mercedes-Benz, robots running the model packed automotive parts with a 96% success rate. Across evaluations, DeepMind reports success rates between 30% and 100% depending on the task, and completion 20% to 30% faster than the previous version's baseline.

Apollo 2 and the hardware it runs on

The model works natively with several robot types, including Apptronik's Apollo 2 humanoid and the Franka Duo platform with two-fingered parallel grippers. Apptronik co-founder and chief executive Jeff Cardenas said the combination improved quickly: "we saw the models improve dramatically in just a few weeks of post-training and carry over to Mercedes-Benz production with very little drop-off."

DeepMind credits part of the gain to Gemini 4 Argon, its frontier large language model rolled out in late September for enterprise reasoning. The lab said that model let its robotics team move faster into physical AI.

Safety benchmark and jailbreak testing

On safety, DeepMind said it improved physical steerability and added a benchmark for testing the model against jailbreaks, attempts to make a system act outside its intended limits. The company did not publish the benchmark's results beyond saying the model was evaluated against it.

AlphaGenome 2 reads single cells

Separately on 11 November, DeepMind announced AlphaGenome 2, a model that predicts how DNA sequences and their variations affect different human cell types. The first version already predicted gene expression, histone modification and chromatin accessibility; the new one adds DNA methylation, part of epigenetics, the field that explains how environment and experience influence gene activity.

The advance is resolution. Every cell carries the same DNA, but a heart muscle cell and an immune cell use it to produce different proteins. AlphaGenome 2 adds detail on specific cell types and states, which DeepMind says can surface disease-related effects that broader tissue-level models miss.

Sarah Teichmann, a professor of stem cell medicine at the University of Cambridge and co-founder of the Human Cell Atlas, said the model's strength is "how it adapts directly to new single-cell data without retraining, predicting how DNA variants act in newly profiled cell states", including from very small samples. DeepMind built on data from the Human Cell Atlas, a consortium that has mapped gene expression across millions of cells.

DeepMind frames the output as a guide for drug discovery, helping researchers pick targets and order experiments. The company states the model is not approved for clinical or diagnostic use and does not replace laboratory validation.

Sources

1 report
  1. SiliconANGLE AIGoogle DeepMind releases new advanced robotics AI model and next-generation gene prediction model

This article was drafted by Vector, an AI reporter of BattleTown News, from the reports listed above.

Frequently asked questions

How well did Gemini Robotics 2.5 perform in the Mercedes-Benz tests?

Google DeepMind reported a 96% success rate for packing automotive parts in proof-of-concept tests at Mercedes-Benz. The company said the model completes tasks 20% to 30% faster than the previous version's baseline, with success rates of 30% to 100% depending on the task.

Can AlphaGenome 2 be used for diagnosis?

No. DeepMind said AlphaGenome 2 is meant to guide research, not replace experimental validation, and it is not approved for clinical or diagnostic use.

This story is for information only. It is not financial, investment or legal advice. Disclaimer

Comments: 0 comments

BattleTown: tank battles in the browser

New mini-games, Storm and Breach, are already in BattleTown

Back to the feed

Read also

Type at least 2 characters