Google DeepMind introduces a new robot-control model, Gemini Robotics 2.5
DeepMind reported a 96% success rate for Apptronik's Apollo 2 kitting car parts in Mercedes-Benz plants, and 30% to 100% higher success than earlier baselines across 200-plus tasks.
- Models & capabilities
- Minor
Google DeepMind introduced Gemini Robotics 2.5, which it calls its most capable vision-language-action model, one that turns camera input and instructions into robot movements. It is rolling out to robotics partners; the post does not describe general public access.
DeepMind said it tested the model on more than 200 distinct tasks, among them serving food with tongs, tying a bow on a gift box and pipetting liquid. Against previous baselines, which are its own earlier models, it reported success rates 30% to 100% higher depending on task type, and tasks completed 20% to 30% faster. The model was shown on several robots, including Apptronik’s Apollo 2 humanoid and the Franka Duo platform.
The most specific industrial figure is a proof of concept on active Mercedes-Benz production lines in Kecskemét, Hungary, and Sindelfingen, Germany. Apollo 2 picked metal brackets from bulk bins and placed them in trays in the order each vehicle needs, a task known as kitting. DeepMind reported a 96% success rate. It describes the work as a pilot, and the post does not give the number of trials or test conditions.
On safety, DeepMind said inference-time reasoning guidance improved safe task completion by 33% around obstacles and 43% around nearby people in its own evaluations. It also released ASIMOV-Jailbreak, an open-source benchmark on Hugging Face of prompts that disguise unsafe requests, and reported that Gemini 4 Argon resisted 99.5% of them against 82% for the earlier Gemini Robotics ER 2.
DeepMind also credited Gemini 4 Argon with speeding development, using it to generate simulated practice runs from real camera scenes and to analyse video of failures. All of the figures are from DeepMind’s own evaluations, with no independent comparison. The release follows Gemini Robotics 2 in July.