Unitree R1 Robot

How much does the Unitree R1 cost in 2026?

Unitree's own online store, checked 30 August 2026, lists three biped configurations: R1 AIR at $4,900, R1 at $5,900, and R1 EDU at $10,500. All six colour variants showed as in stock. The Chinese price is 29,900 yuan after a 10,000 yuan cut on 24 June 2026, down from a 39,900 yuan launch price set in July 2025.

The $10,500 EDU figure is worth pausing on, because it is the single most misreported number in this market. Unitree's specification page still says contact sales for the EDU, and a lot of catalogue sites still print a rumoured figure of $35,000. The store itself lists $10,500 for the EDU Standard Edition. Anything above that from a reseller is a dealer-configured bundle — hands, an Orin compute module, training, support — not Unitree's price.

What separates the tiers is not cosmetic. The AIR has 20 degrees of freedom, four per arm, no waist joints, no head joints, and a monocular camera. The standard R1 has 26 DOF, five per arm, two in the waist and two in the head, and a binocular camera. The EDU has 26 to 40 DOF, keeps the binocular camera, and is the only one that accepts dexterous hands, an optional NVIDIA Orin module at 40 to 100 TOPS, and the SDK. Warranty scales too: six months on the AIR, eight on the R1, twelve on the EDU including secondary-development support.

One genuine cheaper option exists and is often confused with the biped. The R1-D family, announced 30 April 2026 from 26,900 yuan and listed from $4,290, is a dual-arm mobile manipulator on a fixed or wheeled base — it is not a bipedal humanoid, and every variant showed out of stock on 30 August 2026.

What does the entry price actually exclude?

Four things, and together they are the difference between a toy and a platform.

The SDK. On the $4,900 and $5,900 SKUs there is none. Unitree markets the R1 with a line about fully open control interfaces for joints and sensors, which is true of the EDU and misleading on the cheaper models. If you intend to write a single line of code for this robot, your entry price is $10,500, not $4,900.

Hands. Fixed fists on AIR and standard. Dexterous hands are an EDU option and Unitree does not publish an accessory price list — you email sales. Reseller bundles with five-finger or tactile hands sit in the high twenties to low thirties of thousands, which tells you the order of magnitude even if it is not a Unitree price.

Batteries. One 199 Wh pack in the box, roughly an hour of runtime. A spare is $500 on Unitree's store, listed as a standalone item since 4 June 2026. Any serious session needs two or three, so add $500 to $1,000.

Getting it to you. Unitree quotes shipping between $300 and $1,200 and states plainly that the customer is responsible for all customs duties, taxes and import clearance. For a US buyer in 2026 that has meant roughly 2.5% base duty plus a 25% Section 301 tariff on a Chinese-origin robot, which is why owner reports cluster around a 27% uplift. A $5,900 R1 self-imported lands nearer $8,000 to $9,000 all in — which is almost exactly what US dealers charge at $8,990, only they handle the paperwork and give you someone to call.

How autonomous is the R1 really?

As shipped, essentially not at all, and this deserves to be said clearly because the marketing sits right on the edge of it.

The athletic routines are real reinforcement-learning motion policies running on real hardware. That is not the question. The question is who decides to run them, and the answer is a person with a controller. The R1 does not perceive a task, plan it, and execute it. It has no task autonomy, no manipulation autonomy, and on the AIR and standard SKUs no route to acquiring any, because you cannot install software on it.

The onboard multimodal model handles conversation and image understanding — you can ask it what it sees. That is interaction, not agency. There is a real distance between a robot that can describe a cup and a robot that can pick one up, and the R1 does not cross it without an EDU, dexterous hands, and somebody's months of work.

Unitree's own founder Wang Xingxing put the industry ceiling more bluntly than most vendors would, telling CNBC on 19 August 2026 that the ChatGPT moment for humanoid robots could be ten years away. He was talking about the category, not the R1 specifically, but the message applies. Buy this machine for what it does today, in a lab or a classroom, with a competent person driving it. Do not buy it against a roadmap.

What can you actually build on the R1 EDU?

Quite a lot, provided you accept where the documentation stops.

The main SDK is unitree_sdk2, written in C++ and published under the permissive BSD-3-Clause licence. As of 30 August 2026 it contains a dedicated R1 example directory with audio, high-level and low-level subfolders, which is the clearest signal that the R1 is a first-class target rather than an afterthought. There is a Python wrapper with a matching interface, pinned to a specific CycloneDDS release. Because the transport underneath is CycloneDDS, ROS 2 messages can talk to the robot directly without wrapping the SDK, and Unitree publishes a ROS 2 repository for that — though its README currently lists Go2, B2 and H1, not the R1.

Two control levels are exposed. High-level gives you task-space control — locomotion mode, posture, velocity — with the balancing safeguards left on. Low-level gives you per-joint position, velocity, torque, stiffness and damping, with no safety overrides at all, which is exactly what a research group wants and exactly what will break a robot in the hands of an undergraduate. The robot lives on a 192.168.123.x subnet over gigabit ethernet and your development machine has to join it.

The honest gap is simulation. Unitree maintains MuJoCo, Isaac Lab and Isaac Gym repositories, and none of their READMEs list the R1 — they cover Go2, H1, H1-2 and G1. A model repository reportedly includes R1 description files, but sim-to-real for this specific robot does not look first-class yet. If your project depends on training in simulation and deploying to hardware, verify that before you order.

The wider ecosystem is healthier than the R1-specific tooling: an XR teleoperation project, a LeRobot fork for imitation learning, world-model and VLA repositories, and an app-template repository all sit under Unitree's GitHub organisation with meaningful star counts.

Is the Unitree R1 a security risk?

This is the section that changed most in the last week of August 2026, and the accurate answer is more nuanced than either the headlines or the vendor would like.

On 27 August 2026 the researcher Olivier Laflamme published two vulnerabilities in the Unitree G1 EDU, tracked as CVE-2026-76639 and CVE-2026-76640, both giving root. The first is a path traversal in a knowledge-base upload workflow that writes a file into a whitelist directory which is then executed as root. The second is worse in practice: a Bluetooth GATT characteristic accepts writes with no pairing and no authentication, and a Unitree cloud endpoint would hand back a robot's key material to any logged-in account without checking that the account owned that robot. Chained with a buffer overflow in the Bluetooth server, that gives unauthenticated root code execution on any G1 within Bluetooth range. Researchers describe the result as wormable — a compromised robot can attack other robots nearby with no extra setup.

Whether the R1 is affected is genuinely unresolved. Laflamme confirmed the flaws on G1 firmware only, reproduced on four G1 units, and explicitly declined to claim that similarity of software stack proves every Go2, B2 or R1 build is vulnerable. That restraint is worth respecting. The only reporting that names the R1 alongside the vulnerable models is secondary coverage of the earlier UniPwn research from September 2025, which found a single hardcoded AES key shared across units in the Bluetooth Wi-Fi provisioning flow.

Unitree's response was faster than its reputation suggests: the cloud ownership check was patched by 6 August 2026 and a $5,000 bounty was paid, split $4,000 for the Bluetooth chain and $1,000 for the other. As of 29 August 2026 no fixed firmware release had been confirmed for the on-device flaws.

Separately, security reporting and an academic preprint have described G1 units transmitting telemetry to servers in China at short intervals, with fallback channels including a mesh through other Unitree devices on the same network. That reporting is secondary and Unitree has not confirmed it, but it is the substance behind the regulatory action described below. Practical advice regardless of model: put the robot on an isolated VLAN, keep Bluetooth off when not provisioning, and do not run it on a network that touches anything you care about.

Is the Unitree R1 banned in the US?

No — and this is one of the most confidently wrong claims circulating right now.

On 28 July 2026 the FCC released Public Notice DA 26-786, adding foreign-produced advanced robotic devices to its Covered List, with press coverage running 28 to 30 July. Chairman Brendan Carr framed it as securing critical supply chains, citing supply-chain and cybersecurity risk. The rule is country-neutral on its face: "foreign-produced" means failing the Buy American domestic-end-product test — US manufacture with more than 65% of component cost domestic — so European, Japanese and Korean machines are caught by the same wording. The scope reaches mobile ground robots above roughly 2 kg that carry environmental sensing, some autonomy and connectivity above 200 kbps — humanoids, quadruped robot dogs and warehouse AMRs — and the same action covered power inverters. Fixed industrial arms and medical devices are outside it.

The crucial detail is the mechanism. It blocks new FCC equipment authorisations. It is not a customs ban and it is not a recall: it does not revoke existing grants, it does not stop continued use of devices already in the country, it does not stop the sale or resale of models that were already authorised, and a companion order guarantees that authorised devices keep receiving software and firmware updates through at least 1 January 2029. Unitree said as much in its own IPO prospectus filed 31 July 2026, naming the G1, H2 and R1 humanoids and the Go2, B2 and A2 quadrupeds as models that may continue to be sold in the US, while flagging the risk that future models could be barred and that existing authorisations could in principle be revoked later.

So an R1 bought today is legal to import, own and use in the US. What is uncertain is next year's model. Unitree also disclosed that its US revenue share has fallen to 13.30% from around 18-19% in earlier periods, and the Pentagon has added the company to its list of firms with ties to the Chinese military — which matters if your institution takes federal money and has a procurement policy that follows those lists. Check that before you order, not after.

Where are R1 units actually working today?

In classrooms and research departments, almost exclusively.

The best-documented deployment in North America is St. Clair College in Windsor, Ontario, which announced on 21 July 2026 that it was the first college on the continent to take delivery of an R1. The unit went to the Research and Innovation Department, funded through an NSERC Applied Research Tools and Instruments grant and purchased via the Canadian robotics firm InDro Robotics. The stated uses are process monitoring, quality inspection, materials handling, workforce training and predictive maintenance. Local coverage on 23 July 2026 put the cost at about $26,000, which is well above Unitree's list and consistent with a dealer-configured EDU rather than a store order.

The largest single order made public is educational too. On 24 July 2026 the US coding-education franchise iCode announced R1 EDU units going to 16 campuses for an eight-week mentor-led programme aimed at high-school students, with the specification quoted at 24 to 40 DOF and a 100 TOPS Jetson Orin module.

What is missing from the record is any named commercial or industrial company running R1s in production. We looked and did not find one. That is not damning — the robot has only been shipping since around April 2026 — but it does tell you what this machine is currently for.

The market context supports the same reading. Unitree stated in January 2026 that it shipped more than 5,500 humanoids in 2025 out of more than 6,500 produced, and that approximately 70% went to universities and research institutions. Education is not a side channel for this company. It is the business.

Unitree R1 vs Booster K1 and the rest of the sub-$10k field

The R1's real competitor is not the Tesla Optimus or anything with a Boston Dynamics badge. It is Booster Robotics' K1, which starts at $5,999 and — this is the whole argument — ships with SDK access at every tier. The K1 is smaller at 95 cm and 19.5 kg, and it won the KidSize division of RoboCup 2025, so its locomotion is competition-tested rather than demo-tested. If your budget is around six thousand dollars and you intend to program the robot, the K1 gives you that for $99 more than an R1 you cannot program.

Unitree's answer within its own range is to move you up. The G1 is $13,500 on the store, but the G1 EDU with SDK runs $43,900 or more at US distributors. Against that, the R1 EDU at $10,500 is the cheapest programmable biped Unitree sells, and it is genuinely competitive.

Further up, Unitree's store lists the H2 at $29,900, the H2 Plus at $100,000 and the older H1 at $90,000. EngineAI's PM01 is around 188,000 yuan. Booster's T1 sits somewhere between $22,000 and $75,000 depending on which source you believe — we could not resolve that and would not quote a figure.

One headline worth defusing: Forbes reported on 18 August 2026 that a San Francisco company, Nori Robotics, launched a US-made robot at $1,688 with delivery this autumn. It is real, it began shipping units in July 2026, and it has 19 degrees of freedom and a 55 kg lift. It also has a wheeled base, so it is not a bipedal humanoid and does not compete with the R1 on the thing the R1 is for.

What does owning an R1 actually cost over a year?

Start with the robot at list. Add $300 to $1,200 shipping from Unitree, plus duty and clearance if you are importing yourself, which in the US has been running around 27% on Chinese-origin robots. Add one or two spare batteries at $500 each, because an hour of runtime is not a working day. That is your first invoice, and for a US buyer on a standard R1 it lands somewhere near $9,000 to $10,000.

Then the parts nobody quotes. Warranty is six, eight or twelve months by tier — short by Western industrial standards. Unitree does not accept returns. There is no published out-of-warranty service rate, no published spare-part price list for limbs or joints, and no published repair turnaround; we could not verify any of these, and buyers outside China should assume parts availability is a genuine unknown rather than a formality. Individual motors are listed on the store — the A1 motor at $500, a smaller unit at $369 — which at least suggests some component-level supply exists.

Then the largest cost, which is people. An R1 EDU is a research platform; every review that has actually used one says the same thing, that it needs an operator with a technical background and a clear deployment plan. The only structured figure available anywhere is iCode's eight-week mentor-led curriculum for high-school students, which gives you a sense of the ramp for a taught programme rather than a research one.

Budget honestly: robot plus roughly 60-80% again in the first year once import, spares, and someone's time are counted. If that number does not work, the cheaper SKU will not rescue it — it will just give you a robot that does less.

Is Unitree a stable company to buy from?

More stable than it was, and more volatile than the price chart of the last two weeks suggests.

Unitree listed on the Shanghai Stock Exchange's STAR Market on 19 August 2026 under ticker 688836, having priced at 150.80 yuan a share for a valuation around 61 billion yuan — roughly $9 billion — and raised about 6.1 billion yuan. Retail demand was extraordinary, oversubscribed on the order of 8,000 times. The stock opened at 1,100 yuan, up 629%, and closed the first day at 845 yuan, up 460%.

Then it fell. It dropped 18.7% on 20 August, and by 25 August had touched 588 yuan intraday — down 45 to 48% from the debut peak, with roughly 200 billion yuan of market value erased. Trailing P/E at that point was around 877 times against a STAR Market industry average near 38.6, and Nomura's target price of 370 yuan implied a further 40% fall. Chinese state media called it rational repricing; Western wires called it a bubble deflating. Both descriptions fit.

The underlying business is real and growing, with one number that should give a buyer pause. 2025 revenue was about 1.70 billion yuan, up 335%, with net profit around 278 million yuan and gross margin near 60%. But first-half 2026 revenue of 1.152 billion yuan, up 48.5%, came with core net profit down 19.3% year on year. Growth is decelerating and margin is being spent. For a customer, the relevant reading is that Unitree is very well capitalised for the next few years and under real pressure to keep shipping new models — which is good for your roadmap and bad for the resale value of the one you buy.

What is the Unitree Superman, and does it affect the R1?

On 16 August 2026, three days before its Shanghai debut, Unitree unveiled a robot it called Superman, claiming a top speed of 12.66 m/s and a two-metre standing vertical jump on legs only 0.85 m long. The company said it was developed in a little over three months and that the hardware still has significant room for improvement.

The comparison everyone reached for is Usain Bolt, whose peak speed during his 2009 world record was about 12.2 m/s. That comparison does not hold up. Every Superman figure comes from a company demonstration video, no third party has measured it, no source reports the robot completing a 100 m run, and matching a robot's instantaneous peak against a human's peak inside a race is not a like-for-like measurement. Treat it as an unverified capability demo timed for an IPO week.

For an R1 buyer the practical answer is: it changes nothing. Superman has no announced price, no availability and no SKU, and no source establishes whether it shares the R1's actuators, chassis or control stack. It is not an R1 variant, and nothing about it should factor into a purchase decision this year.

Who should buy an R1, and who should walk away?

Buy the R1 EDU at $10,500 if you are a university lab, a college programme, a school district running a serious robotics curriculum, or a company that wants a real bipedal platform for locomotion research or teleoperation work and has someone on staff who writes C++ or Python and is comfortable on a Linux network stack. At that price, with BSD-licensed SDKs and an active repository, it is arguably the best value in programmable bipedal hardware, and the deployment record — a Canadian college, sixteen US campuses — is exactly this profile.

Buy the R1 AIR at $4,900 or the standard R1 at $5,900 only if you want a demonstration object. Trade shows, open days, filming, a museum piece, a very expensive way to make a room go quiet. Those are legitimate, and the machine is genuinely impressive at it.

Don't buy any R1 if you want a robot that does something useful around a house or a warehouse; it cannot pick anything up on the cheap SKUs and has no task autonomy on any of them. Don't buy the cheap SKUs if you plan to program later — there is no upgrade path, you would be buying the EDU twice. Don't buy at all if your institution's procurement rules track federal restricted-vendor lists, without checking first. And if you want a programmable biped for under $6,000, look hard at Booster's K1 before you commit; on that specific requirement it beats the R1 on the only axis that matters.