Atlas GR3 Hand Has 13 Degrees of Freedom, No Pinky and a 100,000-a-Year Target

Boston Dynamics' new Atlas hand has 13 degrees of freedom, up from 7, and the company's target is 100,000 hands a year, according to IEEE Spectrum's Evan Ackerman. No price, no production start and no customer appear in any of the three accounts reviewed here.

The number that matters is not 13 but the count of parts that can fail. By the company's own arithmetic, every added finger adds actuators, cost and breakage risk, and that is why this hand stops at four digits.

What the GR3 hand is, by Boston Dynamics' count

Boston Dynamics' blog post "Robot Hands for Modern AI and Real Work" says the previous hand, GR2, had 7 degrees of freedom and the new GR3 has 13. The thumb has 4 and each of the other three fingers has 3, which is 4 plus 3 times 3.

The post says GR3 is slightly larger than an average human hand and keeps the strength of its predecessor, enough to carry a loaded minifridge of more than 100 lb. It uses one type of actuator, driven directly at each joint, fully enclosed with no cables crossing joints.

The transmission is backdrivable, so joints can sense force and give way on a hard impact instead of damaging the gearbox. Dense pressure sensors cover the fingertips and palm.

Spec	GR2	GR3
Degrees of freedom	7	13
Thumb	Not stated	4
Other fingers	Not stated	3 fingers, 3 each
Payload	Similar to GR3	Loaded minifridge, 100+ lb

1. GR2 (previous hand)7 DOF
2. GR3 (new hand)13 DOF

Source: Boston Dynamics blog, 'Robot Hands for Modern AI and Real Work', accessed 2026-10-20

Why there is no pinky

The company's post says the team taped a pinky to a ring finger for a day to see what they could no longer do, and concluded the extra dexterity was not worth it. A fifth finger would add three actuators, plus cost, volume, power draw and failure risk.

Three actuators for three degrees of freedom implies one actuator per degree, so 13 per hand. That is an inference from the post, not a figure it states.

IEEE Spectrum, in an article dated Oct. 1, 2026 and listed in the sources below, quotes Alberto Rodriguez, Boston Dynamics' director of robot behavior: "Hands are a ruthless design trade-off." Ackerman writes that the hand looks clunky beside the humanlike designs other humanoid makers show, and that Rodriguez argues many of those designs give up reliability and manufacturability and suit research and demos.

The 100,000-hands target and what it implies

Ackerman reports that Boston Dynamics is working on the fine details of the design to reach 100,000 hands a year. If each Atlas carries two hands, that is 50,000 robots' worth before spares; the sources do not say what the spare ratio is.

Robotics 24/7 covered the same hand on Oct. 2, listing tool grasps for drills, torque drivers, grinders, nail guns and welding torches. It adds no deployment status and no Hyundai involvement.

The sources also disagree on wording. IEEE Spectrum describes three fingers plus a thumb, while Boston Dynamics counts four fingers with a 4-DOF thumb. Both describe the same 13-DOF layout.

What is announced and what is shipping

This is an announcement. The IEEE Spectrum article describes work toward mass production with no dates and no customers. Boston Dynamics also plans to use reinforcement learning to find uses for finger splay that goes beyond what a human hand can do, according to Ackerman, which means those motions are a research plan rather than a capability.

For context on how humanoid programmes are being financed, see Mind Robotics' $1 billion round with no product. For a field comparison of lab claims against rated numbers, see Ukraine's hexacopter medevac.

What would change this read

A published per-hand cost, or a mean-time-between-failure figure from a customer site, would turn a design argument into a manufacturing one. Until Boston Dynamics gives a hands-per-year shipping rate with a date, 100,000 is a target and 13 is a count.