Reachy Mini sold $1 million in five days by shipping developers a robot with nothing to do

Five days, $1 million and a robot with no chores to do

Fact On July 9, 2025, Hugging Face opened orders for Reachy Mini, a desktop robot about 28 centimeters tall that it built with Pollen Robotics. The Lite version, which plugs into a laptop, cost $299. The Wireless version, with its own Raspberry Pi and battery, cost $449. Co-founder Thomas Wolf posted the next day that sales were close to $500,000.234

Fact Five days after orders opened, the company said it had passed $1 million in sales, TechCrunch reported on July 16. Wolf described the product on the Equity podcast as "a bit like an empty iPhone": a device bought for what other people will build on it.5

Fact Mike Kalil, an industry blogger, estimated that the first day's sales equaled roughly 1,100 to 1,700 robots depending on the mix of models. The company did not publish unit numbers at the time.4

Our read Most robot launches sell a task: folding laundry, moving totes, patrolling a site. Reachy Mini sold no task at all. It sold access to a platform, priced below what a developer would think twice about. For any robotics team weighing a developer-first launch, this is the cleanest public case of how far that approach goes and where it stops.

Hugging Face bought a hardware maker to get here

Fact Hugging Face acquired Pollen Robotics, a French company founded in 2016 by Matthieu Lapeyre and Pierre Rouanet, in April 2025 for an undisclosed sum. Pollen had raised about €2.5 million before the deal, according to TechCrunch. It brought the Reachy 2 humanoid, and it had already worked with Hugging Face on LeRobot, the open-source robotics library.1

Fact Reachy Mini was announced three months later. The launch blog post lists a six-degree-of-freedom head, full body rotation, two animated antennas, four microphones, a 5W speaker and a wide-angle camera. It shipped with more than 15 behaviors on the Hugging Face Hub and a Python SDK, with a MuJoCo simulator so buyers could write code before the robot arrived.3

Our read The order matters. Hugging Face had a developer audience and a model hub before it had a factory. Buying Pollen gave it hardware engineers and an existing robot line, so the first product could be a small, cheap one that pulled its existing users into robotics instead of a humanoid that needed new customers.

A launch aimed at developers and priced like a gadget

Fact CEO Clément Delangue told TechCrunch the company wanted robots in people's hands quickly, rather than a long preorder with no clear date. The Lite was due to ship the month after launch and the Wireless later in 2025. Both were sold as a kit; the product page now says assembly takes about 2.5 hours and needs no soldering.210

Fact Wolf told the Equity podcast that affordability and accessibility were design principles, and that a low price was meant to make people comfortable with a robot at home. He positioned Reachy Mini as hackable and entertaining, set apart from the chore robots of Figure or 1X.56

Our read The $299 entry price did two jobs. It put the robot in the range of a game console, so a developer could buy one on a card without a budget request. It also lowered expectations. Nobody paying $299 expects a robot to load a dishwasher, which kept the launch out of the autonomy argument that has dogged home humanoids.

The community does the distribution

Fact Hugging Face built the channel before the product. In May 2025, a month before its worldwide hackathon, the LeRobot team said more than 1,000 builders planned to take part across more than 80 local sites. Local hosts, from New York's Betaworks to Munich and Miami, ran their own events on the same weekend.7

Fact On May 6, 2026, Hugging Face opened an app store for Reachy Mini built on Hugging Face Spaces, VentureBeat reported. It launched with more than 200 free apps from more than 150 creators, most of whom had never written robotics code. Owners describe a behavior in plain English and an agent writes, tests and packages the app; a browser simulator serves people without a robot.9

Fact By then about 10,000 units had sold since the July 2025 debut, 3,000 of them in the previous two weeks, according to the same report. The product page, accessed in November 2026, lists 500+ one-click apps, 400+ moves and a community of 14,000+ makers.910

Fact The store's mechanics favor reuse. Each app lives as an open-source repository on the Hugging Face Hub, and any app can be forked and changed with further prompts to the agent, VentureBeat reported. Installation is a single button that sends the app to the owner's robot.9

Fact Independent reviewers became a second channel. Jeff Geerling published a hands-on of the $449 Wireless model on February 13, 2026. He praised the build quality and open access, and called it "a robot built for learning."8

Our read Each channel cost Hugging Face little. Local hosts paid for the hackathon venues. Creators filled the catalog for free. A maker reviewer with a large audience assembled and tested it in public. The company's job was to make the robot easy to buy and easy to publish for, and to keep the catalog on infrastructure it already ran.

Cheap hardware, free apps and a price that went up

Fact The store has no revenue model yet. Apps are free, and Delangue told VentureBeat that paid apps might come later because the store runs on Spaces. Hardware is sold direct; the product page says orders ship within 90 days and that buyers pay shipping, duties and taxes.910

Fact Prices have risen. In November 2026 the product page lists the Lite at $399 and the Wireless at $499, up from $299 and $449 at launch. The page also sells add-ons: a $149 hoverboard base for the Wireless model and a $35 NFC accessory pack. The software is released under Apache 2.0.210

Fact The product page now also points buyers to a Reachy Mini mobile app on the App Store and Google Play, and says setup and use can be done entirely from a phone. The Wireless model lists a Raspberry Pi CM4, Wi-Fi, a battery and an accelerometer; the Lite still runs from a Mac, PC or Linux machine over USB.10

Our read A phone app, a motorized base and NFC accessories are consumer features. Together with the higher price, they suggest Pollen is preparing the robot for buyers who will never open a terminal. That is a different audience from the one that built the catalog, and it will judge the robot on setup time and privacy, not on openness.

Our read The business today is a hardware margin plus accessories, with the app store as a demand engine. The price rise suggests demand held up after the early adopters, or that launch pricing was below cost. The public record does not say which. Either way, Hugging Face is still the platform owner that does not charge a platform fee, which keeps the creator side growing and leaves a lever it has not pulled.

What the developer-first launch did not solve

Fact Geerling's review listed real problems. Setup failed until he fixed an IPv6 DNS issue, the desktop app had no Arm Linux build, and many documentation links on a partner's getting-started page led nowhere. The robot's APIs and web interface were open on the network with no authentication by default, and the conversation app connected to OpenAI without asking.8

Fact He concluded it was worth buying to learn robotics, ideally with some Linux experience, but not as a $449 smart speaker. He called it "not an appliance" that delivers instant AI gratification.8

Our read These are the costs of shipping fast to a technical audience. Developers forgive rough setup if the system is open. Parents and schools will not, and Geerling's note that his children shared personal details with the conversation app shows where the next audience will push back. A move from developers to classrooms would need defaults that are locked down, not just documented.

What other robot makers can borrow

Our read The transferable part is the sequence. Hugging Face ran a distributed hackathon before launch, priced the hardware so individuals could buy it, shipped a simulator so people could build before delivery, and then opened a catalog that let non-roboticists publish. Each step widened the pool of people who could add value to the robot without the company hiring them.

Our read The part that does not transfer easily is the existing audience. Few robot startups own a model hub with millions of developers on it. A company without one has to borrow it, through a platform partner, an open-source project people already use, or reviewers whose audiences overlap with its buyers.