Google DeepMind Launches Gemini Robotics 2.5 for General Real-World Dexterity
Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model that achieves 30-100% higher success rates and 20-30% faster task completion across more than 200 dexterous tasks. Apollo 2 humanoids running the model performed automotive kitting at Mercedes-Benz factories with a 96% success rate. Development was accelerated by Gemini 4 Argon acting as a simulation engine, and DeepMind released the open-source ASIMOV-Jailbreak safety benchmark.
Skynet Chance (+0.05 pp): General-purpose embodied agents deployed in real factories expand AI's ability to act in the physical world. This is partly offset by stronger hazard avoidance and 99.5% jailbreak resistance, plus an open safety benchmark.
Skynet Date (1 day sooner): Frontier models accelerating robotics research itself, via simulation and failure analysis, speeds the transition of AI into physical agency.
AGI Progress (+0.04 pp): Large gains in generalist dexterity, cross-embodiment transfer and real-world industrial reliability are significant steps toward what DeepMind calls physical AGI.
AGI Date (1 day sooner): Using frontier AI to overcome the physical-data bottleneck through synthetic digital twins suggests a self-accelerating research loop.
<< All AI news for November 9, 2026
[ Get the daily index digest on Telegram → ]Related AI News
- Gemini Robotics 2.5 Debuts with Factory Trials and Open ASIMOV-Jailbreak Benchmark
- DeepMind's AlphaProof Code Formally Verifies Signal Protocol Implementation
- Google DeepMind Unveils Gemini 4 Carbon with Leading Cybersecurity and Multi-Day Agentic Capabilities
- Google DeepMind Unveils Genie 4 World Model with Audio, Real-Time Steering and Gemini Integration
- DeepMind Announces Genie 4 as Next-Generation Real-Time World Simulation Model