What does a humanoid robot preorder deposit actually buy in 2026?

The $13,500 figure associated with the Unitree G1 is the most accessible humanoid price in the database. The G1 ships. Manufacturer documentation and owner reports from research institutions confirm bipedal locomotion on inclined surfaces, program-based manipulation, and WiFi-controlled teleoperation.

But the G1 is a research platform. Deployment requires Linux and ROS familiarity; manipulation tasks require custom scripting; there is no plug-and-play consumer workflow. The $13,500 buys the hardware. The integration cost is separate, ongoing, and dependent on embedded-systems expertise most households do not have.

The 1X NEO, priced at $20,000 with a subscription model layered on top, occupies a different position. NEO ships, and 1X Technologies has documented early commercial deployments in residential settings. The distinction matters: NEO is the only humanoid in this comparison that has entered real homes at meaningful volume.

What it delivers there is a teleop-dependent assistant, scripted routines for light cleaning and object retrieval, and continuous operator oversight supplied by 1X as part of the service. Per CNET’s coverage of the launch, the $20,000 figure is the hardware entry point. Subscription fees and operator labor are the ongoing cost structure, and 1X has not published a per-month figure for direct consumer access.

NEO’s model is honest about one thing: it prices in the human. The operator is part of the product. That is a coherent commercial structure. What it is not is a path to cost-reducing household automation. If the goal is to spend less on labor by deploying a humanoid, NEO is not that product.

The Level II classification is load-bearing here: a robot that requires continuous human steering to complete tasks is not reducing human labor. It is redistributing it to a remote operator employed by the manufacturer.

This is the core tension in humanoid preorder economics. The robots that ship do so with Level II architecture. The ones promising higher autonomy, and by implication lower labor dependency, have not shipped at consumer scale.

Figure 03 carries a published MSRP of $130,000 in the Robovations database, a figure consistent with commercial institutional deployment rather than household access. Figure AI’s own announcement materials describe production ramp-up for commercial pilots; per TechRadar’s coverage, a consumer SKU has been discussed but not priced or dated.

Figure 03 is in commercial deployment. Apptronik Apollo appeared at CES in 2024 and again in 2025 with consistent engineering credibility but no announced price or ship date. XPENG Iron has XPeng’s automotive manufacturing heritage behind it and demonstrated object manipulation in controlled settings, but no confirmed consumer timeline.

The word “pilot” is doing significant work in humanoid announcements. A pilot is not a product. It is an arrangement between a manufacturer and an institutional partner, typically involving direct manufacturer support, custom task configuration, and ongoing firmware iteration. The economics of a pilot reflect development cost, not production cost. Using pilot deployment as evidence of consumer readiness conflates two different stages of the commercialization pipeline.

Evaluating a humanoid purely on purchase price misses most of the cost structure. Based on documented evidence across the robots compared here, the relevant budget lines are: hardware acquisition, operator or subscription overhead, technical integration and setup, and ongoing maintenance.

• Hardware: $13,500 (G1, research-only) to $130,000 (Figure 03, commercial).
• Operator overhead: Continuous for NEO (remote operator included in subscription); engineer-hours for G1 (scripting and calibration); undisclosed for Apollo and XPENG Iron.
• Integration: G1 requires ROS expertise and pre-mapped environments. NEO requires workflow design and operator training. Apollo and XPENG Iron have not published integration requirements.
• Maintenance: Per the Robovations classification records, all five platforms require regular joint inspection, actuator checks, and firmware updates. G1 documentation specifies monthly IMU recalibration. NEO documents annual professional actuator inspection by 1X.

None of these cost lines appears in a preorder price. That is not unique to robotics. It is, however, a larger gap here than in most consumer categories because the labor and integration costs can dwarf the hardware cost over a two-year deployment window.

NEO’s subscription model is an explicit acknowledgment that Level II humanoids require ongoing human support to function in homes. The model transfers integration and operational risk to the manufacturer.

The trade-off is that the buyer never fully owns the capability: if 1X adjusts pricing, changes the service tier, or discontinues the operator program, the hardware depreciates toward zero usable function. The Robovations classification record for NEO notes this directly as a trade-off: commercial subscription provides training and ongoing optimization but prevents independent ownership and cost reduction.

The G1’s outright purchase model puts integration risk entirely on the buyer. A research institution with ROS engineers can extract genuine value. A buyer without that infrastructure owns a platform that cannot self-deploy. These are not equivalent risk profiles at the same price point, even though the hardware cost difference is only $6,500.

The Apollo and XPENG Iron positions are not yet defined on this axis. Both have demonstrated hardware. Neither has published a business model for consumer deployment. Whether either will follow the subscription path, the outright-purchase path, or a hybrid remains an open question for buyers evaluating where the category is heading.

A preorder price announced before a ship date serves a specific purpose for manufacturers: it establishes market position, tests price sensitivity, and recruits early-access partners. For a buyer, it documents manufacturer intent, not delivered capability. The distinction matters because humanoid robotics is still in a phase where announced capability and verified capability diverge significantly.

Consider what verification would require. For a humanoid to demonstrate Level III performance in a residential setting, it would need to complete a defined set of household tasks end-to-end without operator intervention, in an environment it had not been pre-configured for, across multiple sessions with different conditions. No manufacturer has published a third-party protocol for this kind of verification.

No independent testing organization has released results against a published humanoid task benchmark for the platforms discussed here. The absence of verification infrastructure is not a criticism of any individual manufacturer. It reflects where the category is: early enough that the verification frameworks do not yet exist at scale.

That context shapes how a preorder price should be read. The G1’s $13,500 reflects real, deliverable hardware with documented research-institution use. The $20,000 NEO figure reflects a service model with a verified deployment record in specific residential settings. The $130,000 Figure 03 price reflects industrial-pilot economics.

The unannounced positions of Apollo and XPENG Iron reflect manufacturers that have not yet committed to a commercial model. Each number is internally coherent within its own context. None of them is a consumer product price in the sense that a buyer can purchase, deploy, and operate the robot without significant additional infrastructure.

The economics of humanoid deployment will look different at Level III. A robot that completes tasks end-to-end without operator steering removes the single largest hidden cost in current models: continuous human oversight. That shift changes the subscription calculus, the integration requirements, and the maintenance burden simultaneously. Hardware cost may not fall as fast as the labor overhead around it.

None of the five robots here is close to that threshold by documented evidence. Each relies on teleoperation as either its primary mode or a documented fallback for tasks outside its scripted routines.

Per the Autonomy Ladder framework, teleop dependency as a fallback caps classification at Level II. That cap is not a prediction about where any of these platforms will be in two or three years. It is a statement about what the current preorder price actually purchases.

Buyers evaluating humanoid preorders in 2026 are not purchasing a finished product category. They are purchasing a position in a development pipeline, with hardware as the tangible asset and the promise of autonomy advances as the intangible one.

The hardware price is the one number that is verifiable at the point of sale. The rest of the value proposition depends on software advances that have not yet shipped at the autonomy levels the marketing implies.

A preorder page is a set of claims about a future robot, and the useful discipline is to read it the way a classification record reads a shipping product. What task completions does the company show end to end, without cuts, in environments it does not control?

What does the warranty cover when the robot is a subscription bundle rather than a purchase? Which capabilities are described in the present tense and which in the future tense? The distance between those two tenses is the deployment gap this piece has traced.

None of these questions require technical expertise to ask. They require treating the deposit as what it is: a position on a delivery schedule that no manufacturer in the category has yet hit at consumer scale. A buyer who can wait keeps every option the early depositor pays to hold.