DeepMind Claims 96% Robot Success Rate on Mercedes Factory Lines

By Nathan Brooks

Google DeepMind says its new Gemini Robotics 2.5 model completed automotive parts-handling tasks on Mercedes-Benz production lines with a 96% success rate. The company announced the model on November 10 and is rolling it out to robotics partners only — no public access, pricing, or API documentation accompanied the release.

The headline number comes from a proof-of-concept pilot at Mercedes-Benz plants in Kecskemét, Hungary, and Sindelfingen, Germany. There, Apptronik's Apollo 2 humanoid robots performed automotive kitting: reaching into bulk bins, grasping metal brackets, and placing them into sequencing trays in vehicle-build order.

Two models split the work. Gemini Robotics ER, the embodied-reasoning model, tracked the multi-step workflow, while Gemini Robotics 2.5 controlled the robot's hands and arms. DeepMind reports the 96% success rate for that combined system. What's missing: the post doesn't say how many picks the figure covers or how long the pilot ran. It's labeled a proof of concept, not a production deployment.

What does this change for manufacturers?

Not much yet, and that's the honest read. DeepMind calls Gemini Robotics 2.5 its most capable vision-language-action (VLA) model to date — it takes visual input and language instructions and outputs robot actions. DeepMind says it improves on three axes: dexterity, reliability, and safety. A separate model, Gemini 4 Argon, accelerated development by generating simulated practice runs from real camera scenes.

Three partners are named: Agile, Apptronik, and Mercedes-Benz. DeepMind's blog states the model "is rolling out to our robotics partners." For anyone outside that circle, there's no way to buy, test, or benchmark it.

The lab numbers, and their fine print

DeepMind evaluated the model on what it calls "over 200 distinct tasks" — serving food with tongs, tying a bow on a gift box, pipetting liquid. It reports a 30% to 100% higher success rate than "previous baselines" depending on task type, plus 20% to 30% faster task completion. In plain terms: under the best conditions, the new model succeeded twice as often as the old ones.

But the baselines aren't named, and their absolute scores aren't given. The model runs on Apptronik's Apollo 2, fitted with a 22-degree-of-freedom SharpaWave hand, and on the Franka Duo dual-arm platform with two-finger grippers. Ars Technica reports the system still depends on cloud servers, so the speed gains don't prove low-latency, on-robot operation.

The 4% gap is the hard part

DeepMind director Kanishka Rao told Ars Technica that closing the remaining 4% gap would be a "steep road." Carolina Parada, DeepMind's robotics head, framed the core tension: making a robot excellent at one application conflicts with preserving its ability to handle many different jobs. Pushing for generalization, she said, may mean never reaching 99.9% reliability in every application.

On safety, DeepMind says inference-time reasoning guidance steers the robot around hazards in real time. The company reports this lifted safe task completion by 33% around obstacles and 43% around nearby people.

DeepMind is also releasing ASIMOV-Jailbreak, an open-source benchmark of hundreds of prompts that disguise unsafe requests across dozens of scenes. On it, Gemini 4 Argon scored 99.5% jailbreak resistance versus 82% for Gemini Robotics ER 2. Note the fine print: the 99.5% belongs to Argon, not to Gemini Robotics 2.5. The post gives no ASIMOV-Jailbreak score for the VLA model being released.

What to watch

Every number in this announcement comes from DeepMind or its partners; no independent evaluation is cited. The practical questions ahead:

• Whether outside labs reproduce DeepMind's ASIMOV-Jailbreak scores.
• Whether the Mercedes pilot extends to sustained, shift-long operation with recovery from grasp failures.
• Whether partners beyond the named three get access, and on what terms.
• Whether on-robot compute replaces cloud inference for latency-sensitive tasks.

DeepMind calls Gemini Robotics 2.5 "an important milestone on the path toward solving physical AGI." The factory pilot is concrete progress. But the distance between a 96% proof of concept and a robot that stays on the line through a full shift is, as DeepMind's own researchers acknowledge, a steep road.