Google DeepMind introduces Gemini Robotics 2.5 for more robot tasks
AI

Google DeepMind introduces Gemini Robotics 2.5 for more robot tasks

TechNews Editorial
TechNews EditorialNov 10, 2026 · 1 min read
Share

Why it matters

Google DeepMind’s tests suggest robots could handle a wider range of tasks, while a factory proof of concept showed a 96% parts-sorting success rate.

The facts

  • Google DeepMind says Gemini Robotics 2.5 handles more tasks and completes them faster than earlier models.
  • The company reports a 96% success rate in a Mercedes-Benz factory proof of concept using Apptronik’s Apollo 2.
  • Gemini Robotics 2.5 is rolling out to Google DeepMind’s robotics partners.

Advertisement

Google DeepMind has introduced Gemini Robotics 2.5, a model that controls robot actions based on what a robot sees and the instructions it receives. The company says it can learn a wider range of tasks and complete them faster than earlier models. It is rolling out to robotics partners.

Google DeepMind evaluated the model across 200 distinct tasks, including serving food with tongs, tying a bow and pipetting liquid. It says success rates were 30% to 100% higher than previous baselines, depending on the task, while completion times were 20% to 30% faster.

The model works with different robot types, including Apptronik’s Apollo 2 humanoid and the Franka Duo platform. Apollo 2 has a five-finger SharpaWave hand with 22 degrees of freedom. Franka Duo uses two-fingered grippers.

Read nextGoogle makes Gemini 4 Carbon available to developers

A factory pilot tests parts sorting

Google DeepMind, Apptronik and Mercedes-Benz tested Apollo 2 on active production lines in Kecskemét, Hungary, and Sindelfingen, Germany. In the proof of concept, the robot sorted metal automotive brackets from supply bins into trays in the order needed for vehicles on the line. Google DeepMind reports a 96% success rate.

The robot used Gemini Robotics ER to track the workflow and Gemini Robotics 2.5 to control its hands and arms. Google DeepMind says the system could take on repetitive preparation work while technicians focus on more complex assembly tasks.

A technician fastens a metal bracket inside an unfinished vehicle beside a tray of parts arranged for assembly.
Illustration: AI & Tech News

Google DeepMind reports safety test results

The company says guidance applied while the model runs improved safe task completion by 33% around obstacles and 43% around nearby people in its evaluations. It is also releasing ASIMOV-Jailbreak, an open-source benchmark with hundreds of prompts that disguise unsafe requests across dozens of scenes. Google DeepMind says Gemini 4 Argon resisted 99.5% of those jailbreak attempts, compared with 82% for its previous baseline, Gemini Robotics ER 2.

Google DeepMind says it used Gemini 4 Argon to generate simulated practice runs and help researchers investigate task failures during development. Gemini Robotics 2.5 is rolling out to the company’s robotics partners.

Source: Google DeepMind

Newsletter

Get the best AI & tech news daily

A concise daily digest. Unsubscribe anytime.

We use your email only to send this newsletter.

Keep reading