Gemini Robotics 2.5

What it is

Gemini Robotics 2.5 is a vision-language-action (VLA) model from Google DeepMind. Google DeepMind published it on 10 November 2026 and described it as its most capable VLA model to date ^[1]. A VLA model takes camera images and written instructions as input and outputs actions for a robot, so one model covers both understanding a scene and controlling the robot's movements.

Origins

Google DeepMind said Gemini 4 Argon helped accelerate the robotics research behind the model, including generating simulated practice runs and digital twins from camera scenes. A digital twin is a virtual copy of a real-world setting ^[1].

Key features

Google DeepMind said the model was tested across more than 200 distinct tasks ^[1]. Compared with previous baselines, it reported success rates 30% to 100% higher depending on task type, and tasks completed 20% to 30% faster ^[1]. The model works natively with Apptronik's Apollo 2 humanoid and the Franka Duo platform ^[1].

The release also covers safety. Google DeepMind said inference-time reasoning guidance raised safe task completion by 33% around obstacles and by 43% around nearby people in its evaluations ^[1]. It also released ASIMOV-Jailbreak, an open-source benchmark of hundreds of prompts that disguise unsafe requests across dozens of scenes ^[1]. On that benchmark, Google DeepMind reported 99.5% jailbreak resistance for Gemini 4 Argon, against 82% for the previous baseline, Gemini Robotics ER 2 ^[1].

Position in development

In a proof-of-concept automotive kitting task with Apptronik and Mercedes-Benz, on production lines in Kecskemét, Hungary and Sindelfingen, Germany, Apollo 2 robots reached a 96% success rate, according to Google DeepMind ^[1]. Kitting means gathering the parts a product needs into one set before assembly.

Gemini Robotics 2.5 is rolling out to robotics partners. The named partners are Agile, Apptronik and Mercedes-Benz ^[1].