DeepMind rolls out Gemini Robotics 2.5 after Mercedes pilot
November 10: The Decoder names Kecskemét and Sindelfingen, describes metal-part sorting, and says rollout includes Agile, Apptronik and Mercedes-Benz.
DeepMind is rolling Gemini Robotics 2.5 out to partners after a 96% Mercedes-Benz parts-sorting pilot and an open safety benchmark.
Google DeepMind introduced Gemini Robotics 2.5, a cloud-hosted vision-language-action model it says is 30% to 100% more successful and 20% to 30% faster than earlier baselines on more than 200 tasks. It works with Apptronik Apollo 2 and Franka Duo robots and is rolling out to Agile, Apptronik and Mercedes-Benz. In Mercedes-Benz pilots in Kecskemét and Sindelfingen, Apollo 2 sorted metal parts into trays at 96% success, according to DeepMind. Safe task completion rose 33% near obstacles and 43% near people, DeepMind says. DeepMind also opened ASIMOV-Jailbreak, 1,500 prompts on Hugging Face. DeepMind reports Gemini 4 Argon scored 99.5% versus 82% for Robotics ER 2; Ars Technica reports 98.8% with Argon in the loop versus 73.5% for ER 2 alone.
- Success up 30–100% and speed up 20–30% versus DeepMind baselines on 200+ tasks.
- Mercedes pilots: 96% success sorting metal parts in Kecskemét and Sindelfingen.
- ASIMOV-Jailbreak has 1,500 prompts and is published on Hugging Face.
- Safe completion rose 33% near obstacles and 43% near people.
- Apollo 2 uses a SharpaWave hand with 22 degrees of freedom.
- Public weights are not described; 2.5 still runs on cloud servers.
What each outlet adds
beyond the shared facts- Google DeepMindSharpaWave hand with 22 degrees of freedom; weights not described.
- Ars Technica1,500 prompts on Hugging Face; cloud hosting; 98.8% versus 73.5%.
- The DecoderPlant names, metal-part sorting, and Agile among rollout partners.
- Products
- Apollo 2Franka DuoSharpaWave
- Organizations
- Mercedes-BenzGoogle DeepMindApptronikAgile
Coverage
oldest first · one row per article · UTC- · 1d ago
Google DeepMindFirst report
DeepMind introduces Gemini Robotics 2.5 for robot dexterity
DeepMind reports 30% to 100% higher success, faster tasks, and 96% kitting success on Mercedes-Benz lines.
- · 10h ago
Ars Technica · AI
DeepMind unveils Gemini Robotics 2.5 and ASIMOV-jailbreak benchmark
Apollo 2 robots kitted parts in Mercedes factories at 96% success; the open jailbreak set has 1,500 prompts.
- · 8h ago
The Decoder
Google DeepMind unveils Gemini Robotics 2.5 after Mercedes pilot
DeepMind reports 96 percent success sorting parts and is rolling the model out to Agile, Apptronik and Mercedes-Benz.