DeepMind's Gemini Robotics 2.5 Hits 96% on Mercedes-Benz Line
TL;DR
- All four named hardware partners (Apptronik, Boston Dynamics, Agile Robots, Franka) now route their competitive differentiation through Google's model, narrowing their independent moat.
- The 96% success figure covers a narrow kitting task; DeepMind disclosed no trial count, failure definition, or human-intervention rate, leaving production-readiness unverifiable.
- The open-sourced ASIMOV-Jailbreak benchmark positions Google to define factory-robot safety evaluation standards before regulators or competitors can establish their own.
DeepMind's new Gemini Robotics 2.5 model packed automotive parts at a 96% success rate in a real-world proof-of-concept at Mercedes-Benz, the company said on November 11 in an announcement reported by SiliconAngle that paired the robotics release with a new biology model.
Across more than 200 distinct tasks, DeepMind says the new version achieves a 30% to 100% success rate depending on the task and completes tasks 20% to 30% faster than the previous version's baseline. The range is wide on purpose: the model still misses on some tasks, even as it clears a commercial-grade bar on others.
The vision-language-action model builds on Argon, DeepMind's frontier LLM that rolled out in late September, and is billed as the company's most capable VLA to date for creating an "AI brain" for robotics. It is rolling out via partners Agile Robots SE, Apptronik and Boston Dynamics, and natively supports Apptronik's Apollo 2 humanoid and the Franka Duo platform with two-fingered parallel grippers.
"This is what we built Apollo for: useful work alongside people in real industrial settings," said Jeff Cardenas, Apptronik's co-founder and CEO.
DeepMind released a companion biology model the same day, AlphaGenome 2, which predicts how DNA sequences and their variations affect human cell types and now covers DNA methylation. Sarah Teichmann, co-founder of the Human Cell Atlas and professor of Stem Cell Medicine at the University of Cambridge, singled out "how it adapts directly to new single-cell data without retraining, predicting how DNA variants act in newly profiled cell states, and that it can adapt even from very small samples." The dual drop lands the same day as Google's Gemini Decisions API and healthcare-focused Gemini rollout, one of three DeepMind-stamped launches our tracker caught on November 11.
What others are reporting
-
Google DeepMind Read →
First-party post names all four hardware partners, discloses 99.5% jailbreak resistance, 33% and 43% safety improvements around obstacles and people respectively, and Gemini 4 Argon as the simulation engine.
Gemini Robotics 2.5 increases both the breadth of skills a robot can learn and the speed at which it can acquire them.
-
TechTimes Read →
Only source to flag the missing trial count, failure definition, and human-intervention rate; benchmarks the claim against Bessemer's 99.9% production-reliability threshold and surfaces Mercedes' 2016 automation pivot.
The missing denominator matters to anyone weighing humanoid robots for a factory line.
-
XenoSpectrum Read →
Breaks down the three distinct safety mechanisms (hazard avoidance, instruction resistance, hardware stopping) and explains why their metrics cannot be combined into a single safety score.
Completing a specific task with high probability is not the same as operating stably in a factory over a long period.
-
Crypto Briefing Read →
Frames the open-sourced ASIMOV-Jailbreak benchmark as a potential industry safety standard, lowering barriers for hardware makers while giving regulators and buyers a shared evaluation tool.
The system reached a 96% success rate on pick-and-pack operations in Mercedes-Benz factory trials.
-
The Gadgeteer Read →
Grounds the story in live production deployment at Sindelfingen and Kecskemét plants, noting the predecessor scored zero on tasks Gemini Robotics 2.5 now handles routinely.
Across more than 200 distinct tasks, DeepMind reports success rates running 30% to 100% higher than its previous baselines.
Originally reported by siliconangle.com
Read the original article →Original headline: Google DeepMind Launches Gemini Robotics 2.5, Claims 96% on Mercedes-Benz Automotive Parts Packing