Hugging Face's $399 Open-Source Robot Duck Microduck Tops $2.6 Million in Day-One Orders

In a robotics industry obsessed with million-dollar humanoids and multi-billion-dollar factories, the most successful robot launch of the late summer is a $399 duck. Microduck, a 25-centimeter-tall bipedal robot duck built by Pollen Robotics, the French robotics startup acquired by AI company Hugging Face in 2025, went on sale Aug 27 and immediately became one of the fastest-selling robots ever made. Hugging Face co-founder Thomas Wolf announced on X on Aug 28 that the duck surpassed $2.6 million in orders within 24 hours, selling roughly one unit every four seconds at peak demand, a figure that would put it on track toward the company's stated goal of selling 20,000 units, a number Wolf has described as the potential upper bound for a best-selling robot product in history.

The duck is a deliberate provocation against the industry's prevailing logic. While competitors pour capital into heavy humanoids designed for factories, Microduck is a cheap, failure-tolerant, open-source robot meant to be trained, crashed, repaired and retrained by hobbyists, students and researchers. Its breakout sales are a test of a contrarian thesis: that the future of robotics lies not in a few expensive machines doing dangerous work, but in millions of cheap, community-trained machines learning new skills the way open-source software does.

Key Facts

TechCrunch reported on Aug 27 that Hugging Face unveiled Microduck, describing it as an open-source robot that users can teach new tricks with reinforcement learning, priced at $399 and expected to ship before Christmas. The Register's Aug 27 coverage detailed the hardware, a 25-centimeter, roughly 800-gram biped with 15 motors, a camera, a lidar sensor and two inertial measurement units, with an articulated beak that can pick up small objects. Pollen Robotics' own launch materials said the duck can walk, sit, crouch, get back up from common falls and even roller-skate, and that the full software stack, including robot control, simulation, reinforcement learning and sim-to-real deployment tools, is available on GitHub under an open-source license.

BigGo's finance desk reported on Aug 31 that the duck's first-day performance was extraordinary: within six hours of launch, Wolf had disclosed pre-orders above $1 million, and by the end of the first day the total had passed $2.6 million, with one unit sold every four seconds at peak. New orders placed after that first day now face shipping delays of four to six months, a sign of how far demand has outpaced supply. The hardware is manufactured by Seeed Studio, the Shenzhen-based maker of open-source hardware, which has the capacity to scale production to meet demand that a typical robotics startup could not.

The commercial success is wrapped around a deeper open-source bet. The duck's behaviors are trained in simulation using MuJoCo, a physics simulator, and a reinforcement learning recipe called PPO, then exported to run on the physical robot's neural control loop, a process the robotics field calls sim-to-real transfer. Because the entire stack is open source, any developer can train a new behavior in simulation, deploy it to the physical duck, and share the recipe with the community, which Hugging Face is positioning as the robot equivalent of sharing a model on its platform.

Analysis

The bigger picture here is that Microduck is not really selling a toy; it is selling a distribution strategy for robotics skills. What this really means is that Hugging Face is trying to make physical robot behaviors as shareable and remixable as its machine learning models, and the duck's sales suggest that a surprising number of people are willing to pay $399 to participate in that experiment. The economics are the key to the strategy. At $399, Microduck is cheap enough that failure is affordable, which is essential for reinforcement learning, where robots crash thousands of times while learning. The low price also lowers the barrier to entry so dramatically that the developer base for physical AI could grow from thousands of researchers to hundreds of thousands of hobbyists, and that base is the asset Hugging Face is accumulating.

Thomas Wolf announced on X on Aug 28 that the duck surpassed $2.6 million in orders within 24 hours.

The timing of the launch, coming amid reports that Nvidia is in talks to acquire Hugging Face for around $12.9 billion, adds a strategic layer that neither company has officially confirmed. Business Insider reported, citing a person familiar with the matter, that an agreement has not been finalized, and both companies have declined to comment. But the fit is logical. Nvidia has made robotics a centerpiece of its growth strategy, with its Isaac robotics stack and its GR00T foundation models for robots, and Hugging Face's LeRobot ecosystem is one of the largest repositories of open-source robotics models in the world. A robot that plugs directly into that ecosystem, selling in volumes that prove developers want hands-on hardware, would be exactly the asset a compute giant acquiring an AI community platform would want to own.

There are reasons to be cautious about the headline numbers. Twenty thousand units at $399 is $8 million in revenue, trivial for a company valued at $12.9 billion in reported acquisition talks, so the duck is a strategic play, not a revenue engine. The first-day order surge also reflects novelty and a well-executed launch by a company with a massive developer audience, and the four-to-six-month shipping delay means the true test of demand will be whether those orders are fulfilled without cancellation. And the hardware itself is not open source, only the software, which limits the remixability that the community's most ambitious members might want.

Why It Matters

For the robotics industry, Microduck's success is evidence that the market for cheap, open, failure-tolerant robots is far larger than the industry's focus on expensive humanoids would suggest. It could pull capital and developer attention toward low-cost platforms, which would accelerate the pace of reinforcement learning research by giving thousands of researchers physical hardware to train on. For Hugging Face, the duck is a wedge into physical AI that strengthens its position in any acquisition or partnership negotiation. For developers, it lowers the cost of entry to robotics from thousands of dollars to a few hundred, which could democratize a field that has been the province of well-funded labs.

TechCrunch reported on Aug 27 that Microduck is priced at $399 and expected to ship before Christmas.

There is also a research value to the duck that the consumer framing obscures. For reinforcement learning researchers, a $399 robot that can be crashed without financial pain, repaired by the user, and retrained with an open-source stack removes the single biggest barrier to hands-on physical AI research, which has always been the cost and fragility of the hardware. A graduate student who could never touch a humanoid can now run real sim-to-real experiments on a desktop robot, and if thousands of such students begin publishing results, the collective progress in robot learning could accelerate in ways that the industry's expensive labs, which guard their hardware jealously, cannot match. That is the quiet promise of the open-source duck, and it may turn out to matter more than the sales numbers.

Next Up

In the coming weeks, watch whether the second batch of Microduck orders sustains the early pace, which would determine whether the 20,000-unit target is realistic, and whether the community begins shipping trained behaviors through Hugging Face's platform, which would prove the sharing-economy thesis. The largest near-term question is the reported Nvidia acquisition of Hugging Face: if a deal is announced, Microduck would become part of a compute giant's robotics strategy overnight, and if it falls apart, the duck's ecosystem becomes the company's most tangible bet on physical AI. Either way, the robot duck has already done what few robots have done: it sold more than $2.6 million of hardware in a day and made the open-source robotics community impossible to ignore.