1X's NEO gets 25-DOF hands while autonomy remains the hard part

Why it matters

1X is pushing home humanoids before autonomy is finished. The new hands may remove a real hardware limit, but Expert Mode shows the product still depends on humans for hard tasks.

Bernt Bornich built 1X around a blunt thesis: a humanoid robot has to live around people to become useful, and 1X has now upgraded NEO, its home robot, with tendon-driven hands that bring the hardware closer to the human tasks it has been promising.

The new hands, announced by 1X on July 9, have 25 degrees of freedom: 22 fully actuated degrees of freedom across the fingers and palm, plus 3 at the wrist. 1X says the hands are force-controlled, backdrivable, low-ratio quasi-direct-drive, sealed to IP68, built with tactile sensing, and headed for every NEO the company ships.

That hardware claim is specific. The autonomy claim is narrower. In a July 10 X post, Aligned News framed the announcement around the caveat 1X has had to live with since it began selling the idea of a home robot: some demos show the upper limit of the hardware under human operation, rather than proof that NEO can perform every shown task on its own.

Aligned News on X

WIRED raised the same question and noted that not every clip reflects autonomous execution after asking how the hand videos were made.

For 1X, that distinction is the story. Bornich's company may have taken a serious step toward solving manipulation, the mechanical problem at the end of the robot arm. It has not shown that its AI can reliably turn those fingers into a home worker without humans in the loop.

Bornich's home-first wager

1X says it was founded in 2014 as Halodi Robotics, with Bornich aiming to build general-purpose robots that could coexist with humans. The company first built around EVE, a wheeled industrial humanoid, then made its explicit home pivot in 2023 after deciding that machine intelligence needed to develop in a less controlled setting than factories.

That history matters because 1X has chosen a harder go-to-market path than many humanoid rivals. Tesla's Optimus, Figure, Apptronik, Agility Robotics, Boston Dynamics, Sanctuary AI, and Unitree have all spent significant time showing or placing robots in factories, warehouses, logistics environments, research settings, or industrial demos. 1X is trying to put a robot into the home, where objects are inconsistent, lighting changes, children and pets appear without warning, and a robot may be asked to handle a wine glass, a zipper, a USB-C cable, a grape, or a laundry pile in the same afternoon.

Bornich has been consistent about the target. When 1X announced NEO preorders on October 28, 2025, he described humanoids as crossing from science fiction and research into a product people could touch. 1X offered NEO Early Access for $20,000, with a $499-per-month subscription option scheduled later, and said first U.S. deliveries would begin in 2026.

The new hand is designed to make that pitch less theoretical. 1X says the hand uses tendon drive motors in the forearm, proprietary tendons through the wrist, tactile sensing across fingers and surfaces, and low gear ratios of roughly 5:1 to 15:1. 1X says all 25 degrees of freedom are force-controlled and fully backdrivable, which lets a finger yield to contact and report force back through the same physical path.

Those details explain why 1X is emphasizing hands rather than another whole-body demo. A humanoid with weak grippers can move around a home and still fail at the work that makes a home robot worth paying for. A robot that can sense slip, adjust grip, rotate objects in-hand, and use tools has a better chance of generating the kind of manipulation data that learning systems need.

The hardware claim is stronger than the labor claim

1X's July 9 post highlights small-object manipulation and tool use, not just pick-and-place demos.

1X's order materials still draw a bright line around what buyers should expect. 1X says NEO uses Redwood AI, its generalist AI model, for learning and repeating tasks. The same order page says NEO arrives for early owners with "basic autonomy" and grows over time. For complex tasks NEO does not know, a scheduled 1X Expert can remotely supervise the robot's actions to help it learn and complete the chore.

That turns teleoperation from an embarrassment into part of the product architecture. It also creates the central risk in 1X's plan. Human-in-the-loop operation can help customers get value from an early robot and can feed data back into training. It also introduces labor cost, privacy concerns, scheduling friction, and a different expectation than the phrase "home robot" usually carries.

1X has tried to design around some of that tension. 1X's order page says owners can pilot NEO through a mobile app and VR device, while Expert Mode is scheduled for complex tasks. WIRED reported that 1X says experts can enter only when requested, robot movements can be restricted, faces and sensitive information can be blurred, and the owner can monitor the feed in the app. The point is still unavoidable: a home robot with cameras, remote operation, and unfinished autonomy asks consumers to accept a machine that is both appliance and access point.

The stronger the hand, the clearer that tradeoff becomes. If the new hardware works as described, the bottleneck moves from the fingers to the policy: identifying the task, planning the sequence, handling edge cases, recovering from failure, and knowing when to stop.

1X is building for scale before the proof is complete

The timing is not accidental. On April 30, 2026, 1X announced it opened its NEO Factory in Hayward, California. 1X described the 58,000-square-foot site as a vertically integrated humanoid robot factory, said it had more than 200 employees, and claimed annual capacity of 10,000 NEO units. 1X also said first-year capacity had sold out in five days. That is a company claim, and actual consumer deliveries have not been independently established.

The hand announcement fits that manufacturing push. The numbers are useful only if NEO units reach homes and create a data flywheel. A dexterous robot in a lab is a benchmark machine. A dexterous robot in thousands of homes is a training system, a service business, and a trust problem all at once.

Investors have funded 1X for that larger bet. 1X announced a $23.5 million Series A2 in March 2023 led by the OpenAI Startup Fund, with Tiger Global, Sandwater, Alliance Ventures, and Skagerak Capital participating. EQT Ventures later said it led a $100 million Series B, with Samsung NEXT and Nistad Group also joining, to finance NEO development and its consumer push. Across those two named rounds, 1X has raised at least $123.5 million.

1X has also bought talent around the home-robot problem. In January 2025, 1X acquired Kind Humanoid, bringing in Christoph Kohstall, Kind's founder and a former Stanford scientist and Google robotics team member. That acquisition reinforced 1X's stated belief that humanoids should be developed around people rather than kept in narrow industrial domains until they are polished.

That belief now has better fingers attached to it. NEO's spec sheet lists a 5-foot-6-inch, 66-pound robot with a 154-pound lift rating, 55-pound carry rating, 842 Wh battery, four-hour runtime, NVIDIA Jetson Thor compute, and dual 8.85-megapixel stereo fisheye cameras. The new hand gives that body a more credible interface with the messy physical world.

Bornich's next challenge is proving that interface can be used by software, repeatedly and safely, in houses that do not look like demo sets. The July 9 hand launch gives 1X a stronger hardware story. The market will judge the company on a different question: how quickly those hands can become autonomous labor instead of a dexterous endpoint waiting for a human or an AI model to tell it what to do.