Booster K1 Robot

How much does the Booster K1 cost?

Between about $4,999 and about $12,500 depending on configuration, as of 30 August 2026 — and every one of those figures comes from a distributor, not from Booster.

This is the first thing to understand about K1 pricing. Booster Robotics does not publish an open MSRP page. What exists is a network of authorised resellers in the US, Europe and Asia, each with its own list, its own bundled warranty terms and its own margin. Two sites quoting different numbers for the "same" K1 are usually both right.

The figures we can verify from distributor listings:

- **Entry configuration: from $4,999.** This is the number that put the K1 on every "cheapest humanoid" list. Some US listings show $5,999 as the standing price with $4,999 having been a launch-period offer, so treat $4,999 as the floor rather than the default. - **Fully configured: around $12,500.** US distributors quote this for the top education configuration, typically bundling more onboard compute and a US-serviced warranty. - **Europe: EUR 7,500 excluding VAT, EUR 9,000 including.** That ex-VAT figure is about $8,730 at 0.85889 EUR/USD, the rate on 28 August 2026. - **Education bundles:** quoted, not listed.

What separates the configurations is mostly compute. Reseller listings describe tiers running from roughly 48 TOPS at the entry end up to around 200 TOPS at the top, and some list a Pro configuration at quote-only pricing. Those TOPS numbers are reseller descriptions rather than Booster specifications, and we flag them as such — if the onboard compute matters to your work, get the exact module part number in writing before ordering, because "K1" alone does not determine it.

One more honest note on the headline. $4,999 buys the cheapest walking humanoid worth owning, and it is still a $5,000 research instrument with an 80-minute battery, no hands and no application software. The comparison people should be making is not against a household robot. It is against a used industrial arm or a fleet of quadrupeds.

Can you actually buy a K1 right now?

You can order one. Whether one arrives quickly is a different question, and for a US buyer there is a legal question underneath it.

On availability: the K1 sold out at CES 2026 — every unit on the stand was reserved and delivered against before the show ended — and supply has been tight since. As of 30 August 2026 several US retailers list the K1 out of stock with waitlist or special-order reservation, and quote two-to-four week lead times rather than shipping from stock. That is a change from our June snapshot, which recorded it as buy-now. It is not a supply failure so much as a small company's order book running ahead of a small company's line.

On the legal question: on 28 July 2026 the FCC added foreign-produced advanced robotic devices to its Covered List. The text is country-neutral — a device is domestic only if it is US-built with more than 65% domestic component cost, which catches European and Japanese machines too — but a Chinese-built research humanoid is squarely inside it. The mechanism is narrow: a covered device can no longer obtain new FCC equipment authorization, and without authorization it cannot lawfully be imported, marketed or sold in the US. Anything authorized before 28 July 2026 keeps selling, and the FCC waived its rules so those units keep getting security and firmware updates until at least 1 January 2029. Nobody's existing robot is affected.

We could not confirm from public records whether the K1 holds an FCC grant predating that date. If it does, US sales continue unimpeded. If it does not, new imports need a Department of War conditional approval that has not been reported for this model. That is not a reason to assume the worst — it is a reason to ask. Before paying a US distributor, require the FCC ID for the exact configuration in writing. A legitimate authorised reseller will have it. That single question is worth more than any amount of speculation about the rule.

The RoboCup record, and why it is the most useful thing about this robot

Almost nothing in the humanoid market can be independently verified. There are no consumer ratings worth the name — across the 77 robots we track, exactly one has any public consumer rating at all, and that is a single review. Spec sheets are manufacturer claims. Demo videos are edited. Deployments are announced and rarely quantified.

RoboCup is the exception, and it is why we lean on it here. It is an adversarial competition, run by an independent federation, with published brackets and results, in which teams that did not build the hardware push it for a week straight in front of witnesses.

The record: the K1 won the RoboCup 2025 KidSize Humanoid League, with Booster hardware taking both first and second place. At RoboCup 2026, held in Incheon from 30 June to 6 July 2026 — the first time the event was staged in South Korea — 59 teams competed and 38 ran on Booster robots. In the Middle division the top three were B-Human, HTWK Robots and Rhoban, all on K1s; the Large division went to Tsinghua Hephaestus on a Booster T1. Booster released a statement on 9 July 2026 describing teams on its robots as claiming all championship titles. The RoboCup Federation has also partnered with Booster on discounted pricing for competing teams, and Booster launched a K1 RoboCup Edition around the event.

Two caveats to keep this honest. First, the winners are research teams, and their software is the thing being ranked; the robot is a shared substrate. Second, the federation partnership and the team discounts mean Booster's dominance of the field partly reflects a commercial arrangement, not only technical merit — when a vendor subsidises the hardware for a competition, its hardware becomes what the competition runs on.

Even with both caveats, this is the strongest availble evidence in the compact-humanoid category. A machine that dozens of independent groups can get results out of, in a fixed week, without factory support, is a machine whose documentation and SDK actually work. That is not a small thing to be able to say.

What can the K1 actually do?

Walk, balance, recover, sense and be programmed. That is the honest list, and it is a fuller list than it sounds.

The K1 walks at roughly 0.4 m/s with force-controlled joints and stays upright through contact — which, in the RoboCup context, means being shoved by other robots at speed. It gets itself back on its feet after a fall. Its two-degree-of-freedom head tracks with an RGB-D camera. It hears through a microphone array and speaks through a speaker, and its most viral moment before the soccer results was a dance routine, which is a real demonstration of trajectory tracking even if it looks like a party trick.

What it cannot do is anything requiring hands. Four-DOF arms with simple end effectors will push, gesture and balance; they will not grasp a variable object, and Booster does not claim otherwise. There is no published payload rating for the K1 — the company has not disclosed one, and any number you see quoted is inferred.

Autonomy is where expectations need managing hardest. The K1 ships as a development platform with an SDK and, per reseller descriptions, a no-code interface for simple behaviours. Out of the box it is not an autonomous agent, and there is no published deployment of a K1 doing useful unsupervised work anywhere. Every impressive K1 behaviour you have seen was written by somebody. That is the product working as designed, but it is the opposite of what a buyer reading "embodied AI platform" might assume.

Runtime is the other constraint people underestimate. About 80 minutes on the 5 Ah pack means a working session, not a working day, and the practical answer in every lab that uses these is spare batteries.

Booster K1 vs Unitree G1 vs the rest of the shelf

The K1's only real competitor at this end of the market is the **Unitree G1**, and they are less alike than the price bracket suggests.

The G1 is bigger — 1.32 m, around 35 kg — with far more degrees of freedom, optional dexterous hands, and a much larger public software ecosystem. It is also the most widely owned research humanoid in the world, which cuts both ways: the ecosystem is deeper, and on 29 August 2026 two vulnerabilities were disclosed in the G1 EDU allowing root-level remote code execution, one of them over Bluetooth. That kind of finding lands on the G1 first because that is what security researchers can get their hands on. No comparable audit of the K1 has been published, which is not the same as the K1 being safer.

The K1's case against it is size, price and durability. At 95 cm and 19.5 kg it is a machine a student can carry, drop and reset alone. It is cheaper at entry. And it has the RoboCup evidence, which the G1 does not have in the same form.

Inside Booster's own line the K1 sits at the small end. The **T1** is the larger platform that won RoboCup 2026's Large division; **K2** and **T2** followed, with the T2 launched on 10 July 2026 as a flagship development platform in an NVIDIA Thor-based Professional edition. If you are buying for a lab that will still be running this hardware in 2029, look at the whole line before committing — the T1 is the one with the bigger competitive record above the KidSize class.

Against everything else on the shelf: a K1 costs less than a tenth of an entry Western industrial humanoid and does not pretend to compete with one. Digit is modelled at about $8,500 a month; Apollo is quote-only. Those are logistics machines with service contracts behind them. The K1 is a research instrument, and comparing them on price is a category error.

Who is Booster Robotics?

A Beijing embodied-AI company founded in 2023, reported to have been started by engineers out of Tsinghua's robot control work. Small, young, and — by the standards of this sector — commercially real: reseller material describes more than 700 units shipped to over 200 customers across more than 20 countries, a figure Booster has not itself broken down, and one worth treating as a company-sourced claim rather than an audited number.

The funding record is public in outline. Caixin reported an A+ round of more than CNY 100 million (about $13.9 million) in July 2025, backed by the Beijing Robot Industry Development Fund and the Beijing Artificial Intelligence Industry Investment Fund. In April 2026 the company completed a round of nearly CNY 1 billion — roughly $140 million at prevailing rates. That is a substantial step up, and it lands in a market where Chinese robotics firms are queuing for public listings; Unitree's Shanghai debut on 19 August 2026 opened more than 600% up before sliding into "bubble" coverage, and Booster's peers are all raising into that weather.

What this means for a buyer is mundane but important. Booster is well enough funded to still exist in three years and small enough that your order competes with everyone else's. Support is reseller-mediated outside China. Documentation is in English and demonstrably usable — RoboCup proves that much — but you are not buying an enterprise service relationship, and there is no published SLA, warranty term or spares price list from the manufacturer.

One quiet strength: Booster's strategy is visibly built around getting hardware into the hands of independent researchers, via competition partnerships and aggressive entry pricing. That is the same play that made certain drone and quadruped platforms into standards. It works when the SDK is good, and the evidence says this one is.

What we could not verify

Being specific about the gaps is more useful than filling them.

**Booster's own price list.** There isn't one in public. Every figure on this page is from an authorised distributor, and distributor prices move without announcement.

**The compute tiers.** The 48-to-200 TOPS ladder appears in reseller descriptions, not in a Booster datasheet we could locate. Get the module named in your quote.

**Payload.** No rated figure has ever been published.

**FCC status.** We could not find an equipment-authorization grant for the K1 predating 28 July 2026 in public records. That may mean it does not exist, or it may mean it is filed under a name or configuration we could not match. Ask the seller.

**Unit numbers.** The 700-units-to-200-customers figure is company-sourced.

**Any consumer rating.** There are none, for this or almost any humanoid. If you find a star rating for the K1 online it is a store rating, an unrelated product, or fabricated. The RoboCup brackets are the closest thing to an independent verdict that exists, and that is why this page spends a section on them.