Hugging Face Microduck: Features, Price, Launch Demand and What It Can Do

Microduck is Hugging Face’s $399 entry into affordable physical AI: a small, duck-shaped biped made with its Pollen Robotics team. At 25 centimeters tall and under 800 grams, it is designed to fit on a desk and give developers, students and robotics enthusiasts a physical platform for experimenting with movement, balance, grasping and learned behaviors.

That positioning matters. Microduck is not being presented as a general-purpose domestic helper. Its appeal is that it makes embodied- AI development playful, relatively inexpensive and accessible outside a conventional robotics lab.

What is Microduck?

Microduck is an open-source biped robot with a single eye-like camera and an articulated beak. The beak doubles as a small gripper, allowing the robot to pick up objects without traditional arms. Pollen Robotics describes the platform as something users can train themselves with reinforcement learning, using physics simulation before transferring behaviors to the physical hardware.

The introductory preorder price is $399 before taxes and shipping. Pollen says initial deliveries are targeted for North America, Europe and the United Kingdom before Christmas 2026.

Hardware and core capabilities

For its size, Microduck combines a relatively broad set of mechanical and sensing hardware:

• 15 motors: The degrees of freedom are distributed across its legs, head and neck.
• Vision and depth sensing: The robot includes a camera, a small depth sensor and LiDAR.
• Balance sensing: Two inertial measurement units help track acceleration, rotation and movement.
• Articulated beak: The beak can act as a gripper for picking up small objects.
• Compact form factor: It stands 25 centimeters tall and weighs less than 800 grams.

The advertised behaviors include waddling, sitting, crouching and standing back up after many common falls. It can also pick up objects and perform roller-skating locomotion when the skates are equipped.

The product is therefore built around action rather than only conversation. It can still support audio interaction—the reported hardware includes a microphone and speaker—but its distinguishing features are locomotion, recovery and manipulation.

What can it do out of the box?

Microduck comes with learned behaviors so that it can be used immediately rather than requiring users to train a robot from scratch. Pollen’s published examples include sit-and-stand, grabbing, walking and self-recovery after falling.

Reported demonstrations and launch coverage also show the robot singing or talking and roller-skating. 12 The skating capability is clear, but the available reporting is less clear about the accessory’s precise packaging and whether every preorder includes the skates by default. Pollen’s product description specifically qualifies the feature as available when the skates are equipped.

That distinction is useful for prospective buyers: the robot’s core capabilities are walking, balancing, recovery and basic grasping, while roller-skating is an accessory-dependent behavior rather than evidence that the robot is a fully autonomous all-purpose machine.

The open-source physical-AI proposition

Microduck’s long-term value is supposed to come from what users teach it after purchase. Pollen says behaviors can be learned in a physics simulator, on a user’s machine or through Hugging Face Jobs, then applied to the robot.

This creates a different product model from a closed consumer gadget. Each new behavior can potentially become part of a shared library of physical skills: one developer works on locomotion, another on recovery, and another on object handling. The broader ambition is to make physical-AI experimentation resemble open software development, where tools and learned behaviors can be reused and improved by a community.

The available materials do not establish how large that community will become or how reliably simulation-trained behaviors will transfer to different real-world environments. Those are important questions for evaluating the platform beyond its launch demonstrations.

How successful was the launch commercially?

The first signs were strong, but they should be described as early demand indicators rather than audited commercial results.

Hugging Face Chief Science Officer Thomas Wolf said the company was selling approximately one Microduck every four seconds shortly after launch. 2 In a separate post, Wolf said more than $2.6 million worth of Microducks had been ordered during the first 24 hours. At the $399 introductory price, that figure would imply thousands of orders, although the post is a company executive’s statement and does not provide independently verified sales, revenue or fulfillment data.

The evidence therefore supports a cautious conclusion: Microduck attracted substantial initial preorder interest. It does not yet establish final unit sales, completed deliveries, profitability or sustained demand. The supplied reporting also does not substantiate a confirmed four-to-six-month backlog or a definite loss of the advertised Christmas delivery target.

Microduck versus Reachy Mini

Microduck and Reachy Mini come from the same broader Pollen Robotics and Hugging Face ecosystem, but they target different experiments.

Reachy Mini is an expressive desktop companion focused on human-robot interaction. Its design centers on a movable head and body, with a camera, microphones, speaker and programmable software.

Microduck is the more action-oriented platform. It adds legs, bipedal movement, balance, fall recovery and a beak-gripper. A technical comparison describes Reachy Mini as centered on interaction, while Microduck focuses on motion and physical action.

In simple terms:

• Reachy Mini: better aligned with expressive interaction, speech, vision and desktop applications.
• Microduck: better aligned with locomotion, manipulation, reinforcement learning and embodied experiments.

They are complementary rather than direct substitutes. The choice depends on whether the developer wants to study how a robot communicates or how it moves through and acts on the world.

Why the launch matters for robotics

Microduck reflects a broader attempt to make embodied AI available through smaller, cheaper and more approachable hardware. Instead of asking buyers to start with a full-size humanoid or a research-grade robot, it offers a compact platform that can demonstrate balance, recovery and learned movement on a desk.

Its significance is consequently less about the duck itself than about the development loop it represents: buy a physical platform, train behaviors in simulation, test them in the real world and share what works. The approach could help more developers and students gain hands-on experience with robotics, although the available evidence is still too early to show whether it will produce a durable ecosystem.

The Nvidia connection—and the neutrality question

Microduck launched as Reuters reported that The Information, citing a person familiar with the matter, said Nvidia had agreed to acquire Hugging Face for $12.9 billion. 3 The supplied reporting does not include a joint announcement confirming the transaction, so the acquisition should be treated as a reported deal rather than an independently confirmed closing.

If completed, an acquisition could give Nvidia closer access to Hugging Face’s open-source AI developer community, models and robotics software. That is a strategic possibility, not a confirmed explanation for the reported transaction.

It also raises a potential neutrality question. Hugging Face’s appeal comes partly from serving as a broad, open hub for models, tools and developers. Nvidia is simultaneously a major commercial provider of the computing hardware used across the AI ecosystem. Ownership by Nvidia could strengthen the platform’s robotics and infrastructure resources, but it could also prompt users to ask whether Hugging Face would remain equally neutral toward competing hardware and technology providers.

The bottom line

Microduck is best understood as an affordable open robotics platform with a playful form factor—not as proof that household robots have arrived. Its strongest selling points are the $399 entry price, compact bipedal hardware, built-in recovery and grasping behaviors, and an open reinforcement-learning workflow.

The launch appears to have generated unusually strong early preorder interest, including a company-reported pace of roughly one sale every four seconds. 2 But delivery performance, audited sales and the size of any backlog remain unresolved. For now, Microduck’s real test is whether its early attention turns into a useful shared library of physical-AI skills.