Google DeepMind Launches Gemini Robotics 2.5 With 96% Success Rate at Mercedes-Benz

Google DeepMind Releases Most Capable Robotics AI Model

Google DeepMind announced Gemini Robotics 2.5 on Tuesday, marking a significant advancement in the company's vision-language-action model designed to create an "AI brain" for robotics1

1

. The advanced robotics AI model enhances dexterity, reliability, and safety to help engineers deploy robots in dynamic, unstructured real-world environments where they coordinate alongside humans1

1

. The platform uses multiple computational layers between user instructions and robot movement, with Gemini Embodied Reasoning serving as the vision-language model that plans task completion using instructions and live video2

2

.

Approaching Human-Level Dexterity Across 200+ Tasks

Google DeepMind tested the AI model across over 200 distinct robotic tasks, demonstrating performance approaching human-level dexterity on some real-world applications2

2

. The company evaluated Gemini Robotics 2.5 on activities requiring high degrees of fine motor skills, including serving food with tongs, using a pipette, and manipulating deformable objects2

2

. Google demonstrated Apptronik's Apollo 2 humanoid robots equipped with the model successfully tying a bow with five-fingered hands, a task that has historically challenged robots due to the complexity of handling flexible materials2

2

.

96% Success Rate in Mercedes-Benz Production Tests

The vision-language-action model achieved a 96% success rate packing automotive parts during real-world proof-of-concept tests at Mercedes-Benz facilities1

1

. Robots using Gemini Robotics have already performed work at Mercedes-Benz factories, demonstrating the model's readiness for industrial real-world environments2

2

. In evaluations comparing the new version to its predecessor, Robotics 2.5 achieves a 30% to 100% success rate depending on the task and completes robotic tasks 20% to 30% faster than the previous baseline1

1

.

Gemini 4 Argon Powers Accelerated Learning

Google DeepMind leveraged capabilities from Gemini 4 Argon, the company's frontier artificial intelligence large language model that rolled out in late September, to accelerate robot research and reach into physical AI1

1

. The advanced reasoning capabilities behind Argon allowed Gemini Robotics 2.5 to increase both the breadth and speed with which robots learn new skills and apply them to real-world scenarios1

1

. Apptronik co-founder and CEO Jeff Cardenas noted that "pairing Apollo 2 with Gemini Robotics, 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"1

1

.

Enhanced Safety With ASIMOV-Jailbreak Benchmark

Google DeepMind enhanced safety features with improved physical steerability and introduced ASIMOV-jailbreak, a new open-source safety benchmark for testing robots against potential security vulnerabilities1

1

2

2

. The company is preparing for robots to work alongside people in real industrial settings through this safety benchmark, addressing concerns about deploying autonomous systems in human-occupied spaces2

2

.

Rolling Out Through Major Robotics Partners

Gemini Robotics 2.5 is rolling out today through Google DeepMind's robotics partners, including Agile Robots SE, Apptronik, and Boston Dynamics1

1

. The model natively works with many different types of robots, including Apptronik's Apollo 2 humanoid robots and the Franka Duo platform with two-fingered parallel grippers1

1

. Industry observers will watch how these partnerships translate laboratory capabilities into scalable commercial deployments across manufacturing, logistics, and service industries.

AlphaGenome 2 Advances Gene Prediction for Drug Development

Separately, Google DeepMind announced AlphaGenome 2, a next-generation AI model that predicts how DNA sequences and their variations affect human cell types, creating a powerful biological tool for fighting disease and drug development1

1

. The model adds finer detail about specific cell types and states, potentially revealing disease-related effects that broader tissue-level models miss, though it remains a research tool not approved for clinical or diagnostic use1

1

. Professor Sarah Teichmann, co-founder of the Human Cell Atlas and professor of Stem Cell Medicine at the University of Cambridge, highlighted that AlphaGenome 2's power lies in "how it adapts directly to new single-cell data without retraining, predicting how DNA variants act in newly profiled cell states," even from very small samples1

1

.