1X NEO Hands: 25-DoF Tendon Upgrade Targets Home Dexterity

EDITORIAL / OPINION

The Hands Problem Meets a Concrete Specification

On July 9, 2026, 1X Technologies published detailed specifications for the hands that will ship on every NEO humanoid robot. The new design delivers 25 degrees of freedom—22 fully actuated in the fingers and palm plus three at the wrist—using a tendon-driven architecture with native force control and backdrivability. Every joint functions as a sensor, and fingertip surfaces add high-resolution tactile channels for normal force, contact location, and shear. Peak torques reach 3.5 Nm at the thumb CMC joint and 2.6 Nm at finger MCP joints, with distal flexion forces up to 45 N and wrist torque of 17.75 Nm. Positioning accuracy is stated at ±0.2 mm. The assemblies are IP68 rated and constructed from food-safe materials.

These numbers matter because they directly address the long-standing hardware ceiling that has kept most humanoid platforms from performing the fine, force-sensitive work that dominates household labor. Earlier gripper-style end effectors offered only a handful of verbs. The 1X hand expands the repertoire to include in-hand rotation, precise pinch under load, and real-time slip correction. The company claims the hands have already passed millions of test cycles on components and assemblies, with wrist joints validated beyond two million high-load cycles.

Force Transparency as a Design Principle

Most robot hands rely on high-ratio gearboxes that mask contact forces. 1X instead uses low-ratio tendon transmission—approximately 5:1 to 15:1—mounted in the forearm. This choice keeps distal inertia low and allows external forces to backdrive the fingers safely. When an object begins to slip, shear sensing triggers an immediate re-grasp through the same physical path that delivered the command. The result is a closed-loop system in which every grasp becomes labeled training data without external instrumentation.

The anatomical bias in DoF allocation favors thumb opposition, mirroring human distribution rather than uniform spacing. This layout supports both power grasps for tools and delicate pinch for coins or USB-C connectors. Demonstrated tasks include assembling LEGO, sorting grapes, zipping jackets, pouring from a kettle, and washing the hands themselves. These are not teleoperated sequences; they represent the output of the integrated perception-action loop the company intends to ship.

Production Scale and Vertical Integration

Hundreds of the new hands have already exited a dedicated production line in Hayward, California. 1X manufactures motors, tendons, electronics, and tactile stacks in-house, claiming capacity for 10,000 hands this year. Vertical integration reduces bill-of-materials cost and accelerates iteration, an advantage over modular approaches that source hands separately. The same factory produces complete NEO units, allowing hardware and control software to evolve in lockstep.

Early access pricing stands at $20,000 purchase or $499 monthly lease. Deliveries are targeted for late 2026 in the United States, with international expansion planned for 2027. Pre-order volume reportedly exceeds 10,000 units. These figures place 1X among the first companies to publish both a firm hardware specification and a near-term shipping schedule for a home-focused humanoid.

Remaining Variables: Unsupervised Reliability and Data Volume

The 25-DoF specification narrows the dexterity gap, yet two open questions will determine whether pre-order interest converts to sustained adoption. First, long-duration unsupervised operation in unstructured homes remains unproven at scale. Laboratory cycle counts and controlled demonstrations do not fully capture edge cases such as variable lighting, unexpected object deformation, or cumulative wear on tendon routing over months of daily use. Second, the policy stack that translates tactile and force streams into reliable behavior must be trained on sufficient real-world interaction data. 1X’s World Model Lab aims to close this loop, but the volume and diversity of home-specific data collected before first deliveries will set the practical capability floor.

Competitive Context in 2026

Tesla’s Optimus program has referenced lower DoF counts in public updates, while other entrants continue to iterate on modular or higher-ratio designs. 1X’s emphasis on force transparency and integrated tactile skin differentiates the platform on the dimension most relevant to household chores. Whether competitors match the combination of low-ratio tendon drive, joint-as-sensor architecture, and IP68 food-safe construction before volume shipments begin remains to be seen. The race is no longer solely about leg locomotion or torso compliance; it has shifted to the quality of the physical API at the end of the arm.

Outlook for Home Deployment

If the hands perform to specification under unsupervised conditions, NEO could move from early-access novelty to practical household tool within the first year of deliveries. Tasks that currently require human presence—precise cleaning, small-object retrieval, light assembly—become candidates for automation. The hardware ceiling has been raised; the remaining constraint is the speed at which learned policies can exploit the new sensorimotor bandwidth. That transition will unfold in real homes rather than demonstration videos, and the data collected there will determine whether 1X’s 2026 specification becomes the baseline for the next generation of home humanoids.