1X Gave NEO Human-Level Hands. That Was the Whole Game.

Humanoid robotics spent years selling the wrong miracle. The demos that went viral were about walking — robots climbing stairs, catching themselves on ice, doing backflips. But a robot that can walk into your kitchen and then can't reliably pick up an egg is a very expensive way to stand around. On July 9, 1X shipped the update that actually matters: a new hand for its NEO humanoid with 25 actuated degrees of freedom and something close to genuine human-level dexterity. It's the company's biggest hardware change since launch, and it's aimed at the real bottleneck.

The problem nobody put on the poster

Legs get the attention. Hands get the jobs. Nearly everything a household or a workplace would actually want a robot to do — cook, fold, sort, assemble, tidy, fetch — happens at the end of the arm, in the fine manipulation of objects of wildly varying shape, weight, and fragility. And that is precisely where humanoid robots have been weakest. Most robot hands are stiff, clumsy, and, in the phrase that best captures the failure, "numb" — they can grip but can't feel, closing on objects with the delicacy of a car crusher because they have no idea how hard they're squeezing.

The reason is buried in the gearing. Conventional robot hands use aggressive gear ratios — often 100-to-1 or 200-to-1 — to turn small motors into strong grips. That works for clamping, but it destroys sensitivity: the gearing makes the joint effectively one-way, "write-only," unable to sense the forces pushing back on it. You get strength and lose touch. For picking up a bolt off a factory floor or a grape off a stem, touch is the entire ballgame.

What 1X built instead

NEO's new hand throws out the high-gearing playbook. It uses a tendon-driven system with unusually low gear ratios — between 5-to-1 and 15-to-1 — in which high-torque-density electric motors pull flexible polymer tendons, much the way muscles pull on the tendons in a human hand. The low gearing preserves what 1X calls "force transparency": the joints act simultaneously as motors and as sensors, so the hand can feel the world it's touching and modulate its grip in real time. Add high-resolution tactile sensors on the fingertips — measuring pressure, contact location, and shear — and you get a hand that knows the difference between a kettlebell and a soap bubble.

The capability range is the proof. The same hand can lift a 20-pound kettlebell and pick a grape off its stem without crushing it. It can install a light bulb, pick a single screw up off the floor, fry an egg, and chop vegetables. The joints are rated as backdrivable — you can push them and they give, which is both a safety property and a sensing one — and the wrist has been cycle-tested past two million cycles under load. The whole assembly is IP68-sealed, meaning it's washable and can be immersed in water, so NEO can cook and then clean itself without frying its own actuators.

Home first, on purpose

This hardware is pointed at a specific destination: the living room. NEO is 1X's bet on the consumer home robot — priced at $20,000 outright or $499 a month, with a refundable $200 deposit, and initial US deliveries slated for late 2026. Founder and CEO Bernt Børnich has framed the launch as the moment humanoids stop being a research project and become "a product… something that you and me can reach out and touch."

A home is, counterintuitively, one of the hardest environments in robotics. It's unstructured, cluttered, full of soft and fragile and irregular objects, and shared with people and pets who cannot be fenced off the way they are on a factory line. Strong-but-numb hands are a non-starter there — a robot that can't feel is a robot you can't trust near a wine glass or a toddler. The dexterity and the safety are the same problem, and 1X's tendon-driven, force-transparent hand is an attempt to solve both at once: gentle enough to be safe, sensitive enough to be useful.

The gap that's now software, not hardware

The most honest line in the announcement is that the hardware currently exceeds what the AI can drive. The hand can physically do more than NEO's models yet know how to command it to do — a striking inversion of the usual robotics complaint, where clever software is bottlenecked by crude mechanics. 1X is betting it can close that gap with over-the-air updates, progressively teaching the same physical hand new skills the way a phone gains features. Build the capable body now; let the intelligence catch up over the product's life.

That's a meaningful strategic wager. It means the manipulation ceiling is no longer set by what the motors and tendons can do, but by how fast the learning improves — and manipulation policies are exactly the kind of thing that gets better with fleet data and iteration. 1X says it can produce up to 10,000 of these hands this year on a dedicated line, manufacturing the motors, tendons, electronics, tactile sensors, and soft polymer skin in-house. That vertical integration is what turns a stunning demo into a shippable part.

The humanoid field organized itself for years around the spectacle of locomotion. 1X just made the quieter, more consequential move: it solved the hand. Because when the robot finally arrives in your kitchen, nobody is going to be impressed that it walked there. They're going to hand it a grape.