Humanoid robots 2026: prices, availability and what you can buy

Fully re-verified: 25 August 2026. This is a buying and deployment guide, not a highlight reel of the most impressive humanoid demos. For every platform we separate five questions that Marketing material often blends together: can a buyer actually place an order, is the stated number a price or merely a deposit, where can the robot be delivered, what is genuinely autonomous, and what service model sits behind the hardware?

The market has moved beyond the blanket statement that humanoids are “just prototypes.” Unitree has several models in an official store, 1X takes NEO orders, and enterprise platforms such as Digit, Walker S2, Figure 03 and Apollo 2 are being tested or deployed in real operations. But they still do not belong in one simplistic price ranking. A research platform bought online and an industrial system sold with integration, safety engineering and service are fundamentally different products.

Quick answer: what can you actually buy in 2026?

For research, software development or demonstrations, Unitree currently offers the clearest public buying path. Its official store lists the R1 from $4,900, G1 at $13,500 and full-size H2 at $29,900. EDU configurations and some options are sales-assisted, while duties, taxes and import clearance sit outside the headline price. Source: Unitree official store.

For a home robot, 1X NEO has the most explicit consumer proposition: $499 per month or $20,000 for Early Access ownership, with a fully refundable $200 deposit. 1X says US deliveries start in 2026, with broader markets following later. It also openly describes Scheduled Expert Mode, where difficult tasks can be remotely supervised. That disclosure matters because “home robot” does not currently mean universal, unattended autonomy. Source: 1X NEO order page.

Factories and warehouses use a different buying logic. Digit, Walker S2, Figure 03, Apollo 2 and Atlas are judged by the deployment package: safety, WMS/MES integration, spare parts, interventions, charging strategy, uptime and cost per completed workflow. The absence of an e-commerce button can therefore reflect an enterprise sales model rather than immaturity.

2026 table: price, sales model and real availability

The RoboMorrow class is not a quality score. It measures only commercial accessibility. A means a public price and a real single-unit purchase path; B means pre-order or reservation; C means enterprise sales; D means limited partner/customer deployments; E means announced, roadmap or pre-commercial status.

Why the sticker price is rarely the real humanoid cost

The most common buying mistake is to treat hardware price as total cost of ownership. A lab also needs to budget shipping, VAT and duties, spare batteries, chargers, replacement actuators, hands or grippers, developer access, simulation tools and engineering time. An industrial deployment adds risk assessment, commissioning, safety systems, facility integration, operator training, monitoring, service-level expectations and the labour cost of human interventions.

That is why a low-cost robot is not automatically a low-cost project. A $5,000–$15,000 platform can be excellent for research, control, data collection and demonstrations, yet still be the wrong choice when the task requires an industrial hand, validated safety functions, higher payload, defined uptime and local support. Conversely, an enterprise platform can be excessive for a university project that mainly needs access to sensors and joints.

A — humanoids with a real public purchase path

Unitree R1: the lowest public entry point

R1 is currently one of the simplest ways to put a physical humanoid in a lab without a six-figure budget. Unitree lists it from $4,900, with roughly 29 kg mass, 123 cm height and 20–26 degrees of freedom depending on configuration. Standard variants can be ordered through the store; EDU is sales-assisted. Official R1 page. Buyers should still verify the exact developer interfaces, end effectors and compute included in the chosen version.

Unitree G1: a stronger research and manipulation platform

G1 is listed at $13,500 and is positioned more directly around manipulation and development. Unitree describes 23–43 powered joints depending on configuration, optional dexterous hands and learning workflows based on imitation and reinforcement learning. Official G1 page. For a serious AI lab, the EDU configuration and software access may matter more than saving money on the base hardware.

Unitree H2: full-size hardware with a public price

Unitree lists H2 at $29,900 in its base configuration. The manufacturer states 180 cm height, 31 degrees of freedom and joint torque up to 360 N·m, plus a 2070-TOPS compute platform. EDU remains sales-assisted. Official H2 page. A public price is valuable transparency, but it is not evidence that a specific industrial task is safe or production-ready without a proper application review.

1X NEO: the clearest consumer route

NEO is sold more like an evolving home service than a traditional industrial robot. 1X says early owners receive basic autonomy that grows over time, while Scheduled Expert Mode can provide remote supervision for tasks the system does not yet know. The company publishes $20,000 Early Access ownership and a $499/month subscription. Official 1X information. For European buyers, the key limitation is geography: the first 2026 delivery wave is primarily US-focused.

B/C — reservations and enterprise buying

NEURA 4NE1: a European route matters

NEURA is strategically interesting for EU buyers because product development, commercial support and compliance discussions are much closer to the European market than a direct import from Asia. 4NE1 Gen 3.5 should still be evaluated as a project platform: confirm the hand configuration, safety package, software version, commissioning, delivery window and service model before comparing headline prices.

Agility Digit: the product is the workflow

Digit is a useful example of why enterprise maturity should be measured through operations rather than a retail price. Agility says newer battery capabilities can run for up to four hours and the robot can dock autonomously for charging. Source: Agility Robotics. In logistics the relevant KPI is completed container moves or workflows per shift, intervention rate and uptime—not whether the robot costs ten percent less than another humanoid.

UBTECH Walker S2: energy management becomes part of uptime

Walker S2 stands out for autonomous battery exchange. UBTECH’s annual report describes a roughly three-minute swap, 52 degrees of freedom and load handling up to 15 kg, positioning the system around continuous industrial operation. Source: UBTECH annual report. It illustrates a broader shift: industrial humanoids are starting to be evaluated as infrastructure, where charging, fleet management and maintenance matter as much as dexterity demos.

Figure 03: production is scaling faster than open sales

Figure states that Figure 03 offers 5 hours of runtime, a 20 kg payload, 61 kg mass and 1.2 m/s speed. In April 2026 the company reported delivering more than 350 third-generation robots internally/across its fleet and demonstrating a one-robot-per-hour production cycle at BotQ. Figure 03 specs · production ramp. None of that should be confused with an open retail channel: access remains selective.

In June 2026 Figure 03 returned to BMW Spartanburg for a sequencing workflow. Figure says the prior F.02 deployment had contributed to more than 30,000 vehicles. Source: Figure BMW update. Real factory operation is a strong signal, but an enterprise buyer still needs pricing, service terms, intervention data and application-specific validation.

Apollo 2, Atlas and Fourier GR-3

Apptronik Apollo 2 and Boston Dynamics Atlas are best understood through partner programs and deployment roadmaps rather than public shopping carts. Fourier GR-3 similarly belongs in a project-sales context. For these systems, the purchase conversation should start with the job: environment, payload, cycle time, reach, tool or hand, safety zones and autonomy requirements. A DOF count or stage demonstration cannot answer those questions.

Autonomy vs teleoperation: the most important 2026 filter

A humanoid can complete an impressive task autonomously and still require frequent support in production. Ask for distributions, not highlight videos: how many cycles complete without intervention, what happens when objects move, can the robot recover after a failure, how often does a human step in, and is teleoperation used only for training/emergencies or as part of normal operations?

The economic meaning changes by environment. A home user may accept occasional assistance if the robot still creates meaningful daily value. A factory may lose the entire ROI case if one operator must continuously supervise several robots. RoboMorrow therefore does not assign a simplistic autonomy percentage unless a vendor publishes a clear, comparable methodology.

Buying a humanoid for Poland or the EU: seven checks

• Confirm the exact configuration. Hands, sensors, compute and developer interfaces can change under the same model name.
• Verify EU compliance for that configuration. A Marketing claim is not a substitute for the documentation attached to the delivered system.
• Check battery logistics. Shipping, storage, replacements and end-of-life handling can complicate cross-border projects.
• Ask who owns service. An actuator or hand failure can stop the entire workflow.
• Quantify interventions. Remote support and teleoperation must be included in TCO.
• Run your own acceptance pilot. Vendor demos are discovery material, not an application acceptance test.
• Benchmark against simpler automation. An AMR, cobot or fixed robot may solve a narrower job more cheaply and reliably.

Home vs factory: two very different markets

Home buyers prioritise safety around people and pets, privacy, low noise, simple maintenance and the ability to handle hundreds of small variations. Factory buyers can constrain the task and environment, but demand repeatability, safety documentation, integration and measurable economics. NEO, Digit and Figure 03 are all humanoids, yet they are not direct substitutes.

Which humanoid should you shortlist?

FAQ: humanoid robots in 2026

What is the cheapest humanoid with a public buying path?

Within Unitree’s official channels, R1 starts at $4,900, while R1-D starts at $4,290 through a sales-assisted developer route. Always compare the exact variant because developer access, compute and end effectors vary.

Can I order 1X NEO in Europe today?

1X accepts orders, but says 2026 deliveries are primarily US-focused and broader market expansion follows later. RoboMorrow therefore does not classify NEO as a normally available retail product in Poland at this verification date.

Can a company buy Figure 03?

Figure is producing and deploying Figure 03, but it does not run an open single-unit store or publish a retail price. It remains limited-deployment / enterprise-access rather than buy-now.

Does a public price prove industrial readiness?

No. Price transparency answers whether a transaction can start. Industrial readiness is a separate question involving safety, reliability, service, integration and performance on the actual workflow.

Related RoboMorrow resources

Robot Database · Humanoid market methodology · AMR vs AGV 2026

Sources and update methodology

We prioritise official stores, product pages, regulatory/company documents and deployment updates. We publish a price only when a public source or explicit financing model exists. A deposit is not treated as the full purchase price, production is not equated with deliveries, and a demo is not treated as proof of universal autonomy.

• Unitree official humanoid store
• 1X NEO order page
• Figure 03 official specs
• Figure 03 production ramp
• Figure 03 at BMW
• Agility Digit update
• UBTECH annual report

How to read “availability”: store, pre-order, enterprise and roadmap

In humanoid robotics, “available” can mean at least four different things. A store-listed product can normally be ordered as a single unit. A pre-order means the manufacturer accepts an order or deposit while timing, region and final configuration may still evolve. Enterprise sales require a commercial process, a deployment project and often integration. A roadmap is a product direction, not a purchase path. RoboMorrow keeps these statuses separate because they are fundamentally different for a buyer.

This matters when reading headlines. Starting a production line is not the same as opening public sales, and a deployment with one partner does not mean any company can buy the same system today. In this guide, BUY NOW requires a real purchase route; enterprise means the buyer still has to pass through a commercial and technical deployment process.

Buying for a company: 10 questions before an order

• Does the quoted price include hands/grippers, onboard compute, charger, battery and software licences?
• What spare-parts availability and service response can the supplier provide in Europe?
• Who is the responsible manufacturer, importer or integrator for the EU deployment?
• What safety documentation applies to the exact configuration and workflow?
• How many autonomous task cycles occur between human interventions in the target job?
• Is teleoperation part of normal operation, a fallback, or only a training tool?
• How often does the robot charge, and can batteries be swapped without a long stop?
• Which APIs, SDKs, simulators and data interfaces are available to the integrator?
• Are software updates included and for how long is support committed?
• Can the supplier run a paid pilot on the real workstation before a fleet purchase?

These questions matter more than a polished demo. In an industrial project, a mechanically capable robot can still produce poor economics if interventions are frequent, integration is closed and spare parts travel across continents. Conversely, a platform with fewer degrees of freedom can win if service is stable, the SDK is mature and the workflow is repeatable.

Buying for the home: what to expect in 2026

The home market is earlier than industrial deployment. 1X NEO has the clearest consumer purchasing model, but 1X itself describes expert assistance for tasks the robot cannot yet complete independently. Buyers should therefore ask not only “what can it do?” but also “what happens when it cannot do it?” That distinction is central to understanding a first-generation home humanoid.

In 2026 we do not assume that a general home humanoid can safely cook any meal, supervise a child, climb every staircase and complete every domestic chore without support. A better evaluation starts with a task list: carrying light items, fetching objects, tidying defined surfaces, opening specific cabinets or executing simple routines. Then verify which tasks are autonomous, which need user demonstration and which may need remote help.

Europe and Poland: compliance, service and responsibility

For an EU buyer, the ability to place an online order is only the beginning. Confirm shipping terms, VAT and customs, the responsible economic operator, documentation for the exact configuration, service procedures and lithium-battery transport. For a business deployment, workstation risk assessment and integrator responsibility become part of the project.

RoboMorrow does not treat a CE mark shown in Marketing material as a substitute for the documentation required for the actual unit. Hands, payload, tooling and intended use can change the safety context. If a supplier cannot clearly explain who handles compliance and service in Europe, that is a procurement risk regardless of an attractive hardware price.

Comparing autonomy without a Marketing shortcut

There is no single broadly adopted “autonomy percentage” for humanoids. Measure autonomy per task instead: autonomous cycles between interventions, median time to intervention, grasp success, task-completion rate without help and the range of environments in which the workflow is repeatable. For home use, record how often a person must correct or rescue the robot during the same defined chore.

This avoids two opposite mistakes. Teleoperation is not automatically evidence that a product is fake; it can be a legitimate fallback and a source of training data. But a successful demonstration is also not evidence of universal autonomy. Procurement should be based on repeatability in the buyer’s workflow, not the best clip the manufacturer can publish.

When waiting is the better decision

Delay a purchase when the task is not defined, local service is unclear, the required safety case is not documented, or the business case assumes an autonomy level the supplier cannot measure. A pilot or the next product generation is better than buying “for the future” with no operational owner.

Buying now can make sense for research, development, data collection, education, demonstration and tightly scoped industrial workflows where a pilot is possible. In those cases, technology risk is an explicit part of the project rather than a hidden failure mode in day-to-day operations.

Five buyer scenarios: building a real shortlist

1. University, lab or embodied-AI team

For research, the product is not just the mechanical body. The real asset is development access. Before ordering, verify the SDK, ROS 2 support, raw camera and joint-state access, telemetry rate, logging, simulator availability, licensing for custom models and whether low-level control is permitted. A cheaper base robot can become the expensive choice if the required interface exists only on an EDU edition.

Unitree G1 and H2 EDU, selected EngineAI variants, Fourier GR-3 and NEURA can all make sense in this category for different reasons. R1 lowers the entry cost but should not automatically be treated as a substitute for an education/development configuration. Write down the exact I/O, compute, hand, power and emergency-stop requirements before paying for the hardware.

2. Small or medium company automating its first process

An SME should first prove that the task needs legs. If transport happens on a flat floor and manipulation points are fixed, an AMR with an arm or a cobot may be cheaper, easier to validate and easier to service. A humanoid becomes more compelling when the workplace was designed around people: stairs, doors, varied benches, shelves, carts and tasks that change frequently.

Local integration support can matter more than the purchase price. A platform bought directly without a service path can look attractive in CAPEX and then lose the business case through engineering time, shipping, spare parts and unclear software ownership. A quotation should include commissioning, training, response time, critical spares and responsibility for the integration layer.

3. Large factory or logistics operator

At scale, hardware price is secondary to cost per successful cycle, fleet availability, interventions per thousand operations, MTBF, MTTR, recovery time and the number of robots one person can supervise. Digit, Walker S2, Figure 03 and Apollo 2 should be compared at this operational layer, not by degrees of freedom.

A pilot should use the same containers, parts, lighting, floor conditions and disturbances found on the production shift. The weakest benchmark is a vendor-only demo on ideal objects. The strongest is a repeatable acceptance test with predefined pass criteria and a log of every human takeover.

4. Consumer looking for a home robot

Home humanoids are the least mature market. 1X NEO has the clearest consumer-style commercial path, but early access is not the same as mature home appliance reliability. In a home, the robot meets children, pets, clothing, glass, food, stairs and private documents. Emergency stop, camera privacy, data retention, remote expert access and behaviour after a network loss therefore matter more than a headline walking speed.

If a service uses a remote expert, the buyer should understand when that person can connect, which sensors they can see, where data is processed and whether the user can disable the feature. Teleoperation can be a sensible bridge toward greater autonomy, but its cost and privacy impact belong in the buying decision.

5. Robotics integrator

An integrator can tolerate a less finished platform if it receives stable APIs, documentation and vendor support. Firmware versioning, the ability to freeze a validated release, backward compatibility, log access, configuration backup, rollback and spare-part commitments become critical. Continuous OTA evolution is exciting in a lab but needs change control in production.

How to read a public humanoid price

A starting price is often only the first line of the bill of materials. The base configuration may exclude hands, additional compute, developer access, teleoperation hardware, spare batteries or sensors required by the application. Ask for the complete configuration: robot, hands or gripper, batteries, charger, compute, sensors, control station, licences, fleet software, support, commissioning, training and transport.

Separate CAPEX from annual operating cost. Software subscriptions, connectivity, support, wear parts and teleoperation can reorder two quotations after two years. Include downtime as well: if one hand failure stops the process for a week, a small saving on acquisition price is irrelevant.

What “mass production” actually means

Capacity, run-rate, production and deliveries are different metrics. A line designed for 10,000 robots per year does not prove that 10,000 units are being built. A one-unit-per-hour rate achieved during a particular period does not tell us how many customers received robots. And a manufacturer-owned fleet is not the same as paid external deployments. RoboMorrow keeps these concepts separate because humanoid announcements often blur them.

For a buyer, scaling matters when it improves lead time, spare-part availability and service capacity. A factory that can build hundreds of robots but has no support structure in Europe may be less useful than a smaller supplier with a validated integration path and an enforceable SLA.

Operational metrics we want vendors to publish

Until these metrics are published under comparable protocols, a universal RoboMorrow 9/10 score would be false precision. TOPS, DoF, torque and walking speed describe hardware, but they do not tell a customer how much useful work the system will complete each month.

Acceptance test before signing a fleet contract

• define three to five critical tasks and failure scenarios
• set a minimum task-success rate and maximum intervention rate
• test with the customer’s real objects and surfaces
• measure full cycle time, charging and recovery
• test network loss and software-update behaviour
• validate emergency stop, safety zones and restart
• log every manual intervention
• define re-validation rules for later firmware versions

Only after this pilot should the customer discuss fleet scale. The contract should clearly separate guaranteed performance from roadmap targets and Marketing demonstrations.

Europe: import, service and configuration compliance

For an EU purchase, a CE mark on a website is only a starting point. Verify documentation for the exact ordered configuration, including battery, hands and additional modules. Instructions, responsible economic operator, intended use, cybersecurity, updates and safe deactivation of AI functions all matter. In a workplace, the employer still needs an application-specific risk assessment.

Service is equally important. Ask where the robot physically goes after a failure, which components can be replaced on site, expected lead time for actuators, hands and batteries, and whether an EU partner is trained. Fleet buyers should negotiate critical spares on site and a maximum response time.

How this guide is maintained

This URL is the permanent owner of the buying guide. We do not create a new “Humanoids 2026” page every month because that would split search signals and update history. We refresh the same guide when a public price appears, reservations open, deliveries start, specifications materially change, a major deployment is confirmed or credible production data becomes available. Roadmap claims remain roadmap claims until the commercial milestone actually occurs.