Google DeepMind unveils Gemini Robotics 2.5 AI model

Google DeepMind announced Gemini Robotics 2.5 on November 10, 2026, calling it its most capable vision-language-action model to date, according to Google DeepMind's announcement.

What's new in Gemini Robotics 2.5

• Tested across more than 200 distinct tasks, the model shows 30% to 100% higher success rates than previous baselines and completes tasks 20% to 30% faster.
• In a Mercedes-Benz automotive production-line pilot at sites in Hungary and Germany, it achieved a 96% success rate using Apptronik's Apollo 2 humanoid robots.
• Safety testing shows a 33% improvement in safe task completion around obstacles and a 43% improvement in safe operation near people.
• The model achieves 99.5% jailbreak resistance in safety testing, compared with 82% for the previous baseline.
• It can control complex 22-degree-of-freedom hands and grippers and is compatible with Franka's Duo robot platform.

What it means for AI builders

The way Gemini Robotics 2.5 itself was built points to a broader shift: Google DeepMind used Gemini 4 Argon as a simulation engine to accelerate the model's development. For developers working on embodied or robotics-adjacent AI, that illustrates how frontier large language models are being repurposed as simulation engines to train physical-world agents, not just as conversational tools.

Availability

Gemini Robotics 2.5 is rolling out to robotics partners rather than reaching broad consumer availability, following the pattern set by hardware-dependent pilots such as the Mercedes-Benz production line.

Sources

• Gemini Robotics 2.5 accelerates real-world physical intelligence (Google DeepMind, November 10, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.