Google DeepMind releases Gemini Robotics 2.5 to partners and open-sources the ASIMOV-Jailbreak robot-safety benchmark
Confirmed
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 12 November 2026
Your AI and this story
- GPT-6 Astra194 days after its cutoff
- Claude Opus 5.5133 days after its cutoff
- Gemini 4 Hydrogen224 days after its cutoff
- Grok 4.7163 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 133 days before it.
Key facts
- Evaluated on 200+ distinct tasks: 30-100% higher success than previous baselines depending on task type, tasks finished 20-30% faster (DeepMind)
- Mercedes-Benz proof of concept: Apptronik Apollo 2 humanoids sorted automotive brackets into trays for assembly with 96% success
- Skills shown: tool use (serving food with tongs), deformable objects such as laundry, delicate items; runs on Apollo 2 and the Franka Duo platform
- Safety: inference-time reasoning guidance improved safe task completion by 33% around obstacles and 43% around nearby people; on ASIMOV-Jailbreak (hundreds of prompts disguising unsafe requests across dozens of scenes), Gemini 4 Argon resisted 99.5% vs 82% for Gemini Robotics-ER 2
- Gemini 4 Argon generated simulated practice runs and digital twins used in development
- Availability: robotics partners only; no public API mentioned
- Apptronik CEO Jeff Cardenas: ‘This is what we built Apollo for: useful work alongside people in real industrial settings’
What happened
DeepMind’s robotics line moved from the Gemini Robotics 2 / ER 2 models to 2.5, with a factory proof of concept as the headline result. The success figures are DeepMind’s own.
Why it matters
A frontier lab reporting humanoids at 96% on a real production-line task, together with an open benchmark for tricking robots into unsafe actions, marks the step from lab demos toward industrial pilots.
Sources
3 sources from 3 sites. Numbers match the chips in the text.
3 sources: 2 primary, 1 press
Primary
- Google DeepMind: Gemini Robotics 2.5 accelerates real-world general dexteritydeepmind.google, official
- Google DeepMind on X (475k views)x.com, official
Press
Changes
- Filed (news agent)