How Much Does a Humanoid Robot Cost in 2026? Every Platform from $16K to $150K Compared

Two Completely Different Markets

The Unitree G1 costs $16,000. The Agility Digit costs approximately $10/hour to operate. These are not the same product. They are not comparable. The conflation of these two things in most humanoid robot coverage is why the market is so poorly understood by investors, journalists, and procurement managers alike.

The G1 is a research and development platform that happens to be shaped like a human. You can buy one, program it, run experiments on it, and learn from it. It has no commercially validated task performance in real production environments.

The Digit is a warehouse worker that happens to be shaped like a human. It is not for sale to the general market. It is deployed under commercial contracts in Amazon and GXO facilities, performing specific logistics tasks at defined performance SLAs. It is evaluated on throughput, uptime, and cost per pick — not on research agility.

This article maps every major humanoid platform to the correct market, prices what can be priced, and explains what the RaaS pricing that governs commercial deployment actually means for the workforce economics.

The Accessible Tier: The Most Significant Price Point in Humanoid History

The Unitree G1 at $16,000 is the most significant humanoid robot price point in the industry's history. Not because $16,000 is cheap — it is not cheap for most applications. But because it is the first time a fully bipedal humanoid with functional arms and hands has been available for less than the cost of a used car.

For context:

Unitree G1 — $16,000

What you get: A 35 kg bipedal humanoid with 23 degrees of freedom, capable of walking, climbing stairs, and performing basic manipulation tasks. Unitree's SDK enables custom application development. The G1 is genuinely capable of meaningful research and is increasingly used in academic robotics programmes, AI training data collection, and early commercial task evaluation.

What you don't get: Commercial deployment readiness. The G1 does not come with the task-specific end effectors, safety validation, deployment support, or performance SLAs that commercial operators require. It is a hardware platform, not a deployed solution.

Who should buy it: Robotics researchers, AI training data collection teams, university labs, corporate R&D teams evaluating humanoid task feasibility, developers building applications for the anticipated commercial humanoid market.

Who should not buy it: Operations managers evaluating a replacement for warehouse labour. The G1 is not a commercial warehouse robot. It is a research platform.

Unitree H1 — $90,000

The H1 is Unitree's commercial-grade humanoid — taller, heavier payload capacity, and designed for more demanding physical tasks than the G1. At $90,000, it sits between the research-accessible G1 and the enterprise-contract commercial platforms. Several early commercial deployments are in progress in Asia.

For a direct comparison, see Unitree G1 vs Unitree H1.

The Commercial Deployment Tier: What $10/Hour Actually Means

The commercial humanoid platforms — Agility Digit, Figure F.02, 1X NEO Beta, Apptronik Apollo — are not for general purchase. They are deployed under controlled commercial programmes, priced on RaaS or enterprise contract models, and evaluated on operational performance metrics.

Agility Robotics Digit — ~$10/hour RaaS

Agility is the only humanoid company that has publicly stated a per-hour RaaS price: approximately $10/hour. This is significant because it establishes a reference point for humanoid commercial pricing and enables direct workforce cost comparison.

The workforce arithmetic:

At $10/hour, Digit does not produce a dramatic human-versus-robot cost saving at standard single-shift working hours. The advantage emerges at utilisation that human labour cannot match — 20-hour operating days, weekend operation, no sick days, no turnover cost. The fully-loaded comparison across a full operating year favours Digit at high utilisation.

GXO Logistics — one of the world's largest logistics operators — has deployed Digit in live warehouse environments. This is not a pilot. It is operational commercial deployment.

For a comparison with the other leading commercial platform, see Agility Digit vs Figure F.02.

Figure F.02 — Enterprise Contract, Not Publicly Priced

Figure's F.02 is deployed in BMW's manufacturing operations — a programme that has contributed to over 30,000 cars produced with F.02 involvement. This is the most significant commercial automotive humanoid deployment on record.

Figure does not publish pricing. Enterprise contracts of this nature are negotiated with bespoke terms reflecting task complexity, deployment volume, performance guarantees, and support obligations. The BMW deployment price is confidential.

What the BMW deployment tells us: the F.02 is commercially validated in precision automotive manufacturing — one of the most demanding production environments for consistent quality requirements. That validation is more commercially significant than any hardware spec or price point.

For a side-by-side platform comparison, see Figure F.02 vs Apptronik Apollo.

Apptronik Apollo — Commercial Deployment Programme

Apptronik is in commercial deployment programme phase, having announced partnerships with GXO and others. Pricing is not publicly disclosed. Apollo is built explicitly for logistics and light manufacturing tasks and is designed from the ground up for RaaS commercial deployment economics — not research.

1X NEO Beta — Research Partner Programme

1X Technologies is making NEO Beta available through a research partner programme at no public purchase price. 1X's commercial model emphasises the software and data layer over hardware sales — they are building toward a service model rather than a hardware product model.

Boston Dynamics Atlas — Research and Enterprise Only

The electric Atlas, launched in 2024, is not commercially available for general purchase. Boston Dynamics is developing Atlas as a research and enterprise platform, with Hyundai manufacturing as the primary stated deployment focus. Atlas represents the frontier of humanoid mechanical capability — the locomotion and manipulation performance is not matched by any commercially available platform — but it is not a product you can procure outside a strategic partnership with Boston Dynamics.

For a capability comparison with the most accessible platform, see Boston Dynamics Atlas vs Unitree G1.

The BOM Cost Trajectory: How Humanoid Prices Get to $10,000

Why every commercial humanoid platform costs what it does today, and where prices go from here, is explained by two data points and one economic law.

The data points (ARK Investment Management 2026):

The China figure reflects Unitree's manufacturing efficiency, vertical integration of actuator supply chains, and the scale effects of the Unitree G1/H1 production ramp. The $16,000 G1 retail price is achievable because the underlying BOM has collapsed.

The economic law: Wright's Law — the empirical observation that unit production costs decline by a predictable percentage for every doubling of cumulative production volume — has governed the cost trajectories of solar panels, batteries, and semiconductors for decades. ARK's Wright's Law projection applied to humanoid robots suggests a sub-$10,000 BOM within 8–12 years of mass production.

Mass production, in this context, means annual unit volumes of 100,000+ — a milestone no humanoid manufacturer has reached but that multiple platforms are targeting within 5 years.

For a full treatment of this trajectory, see Wright's Law Applied to Robot Cost Decline.

The Geppetto Catalog Data

Across the humanoid robot platforms tracked in Geppetto's catalog, prices range from $16,000 (Unitree G1, the most accessible humanoid robot ever produced) to enterprise-only contract pricing for commercial deployment platforms. Of the humanoid robots in the catalog with confirmed pricing, the median sits above $80,000 — reflecting that humanoids remain primarily enterprise and research products in 2026. The platforms driving the industry's commercial validation — Agility Digit, Figure F.02, Apptronik Apollo, 1X NEO Beta — are not publicly priced and are not available for general purchase.

What $10/Hour Means for the Workforce

Geppetto's Warehouse Picker/Packer Jobs Index entry scores 88/100 on the automation risk scale. The Digit deployment at GXO is part of why.

The 88/100 reflects a task profile where the dominant activities — picking items from shelves, placing them in containers, moving them between locations — are exactly the tasks that current commercial humanoids are built and validated for. The Deployment Reality component of the score is the most heavily weighted factor, and it is fully loaded: Amazon, GXO, and BMW are not pilots.

At $10/hour running 20 hours/day, a humanoid robot generates the equivalent of 2.5 human-shift equivalents of labour per day. Across a 365-day year, that is 7,300 hours of labour at a cost of $73,000. A human associate working 2,080 hours costs $60,000–70,000 fully loaded — and does not work nights, weekends, or holidays without significant premium.

The crossover economics are real. They are not projections. They are the current operating model at GXO.

For the full workforce analysis, see Best Humanoid Robots 2026 and Humanoid Robots: What They Are and Where They're Headed in 2026.

What You Cannot Buy: The Frontier Platforms

The platforms generating the most investment, press coverage, and commercial interest — Atlas, F.02, 1X NEO, Apollo — are not available for purchase by the general market. This is intentional, not a commercial oversight.

Why commercial humanoid platforms are not for sale:

  1. Deployment quality control. A commercial humanoid that malfunctions in an automotive plant or logistics warehouse creates safety incidents, production stops, and reputational damage. Controlled deployment programmes allow vendors to maintain quality standards, respond to issues immediately, and learn from operational data without the support overhead of thousands of independent owners.
  1. The service is the product. Commercial humanoid companies are not selling hardware. They are selling labour-as-a-service — a defined level of task performance at a defined cost per hour. The hardware is the delivery mechanism. Selling the hardware separately disconnects the performance guarantee from the service obligation.
  1. Data accumulation strategy. Every commercial deployment generates proprietary training data for improving the robot's capabilities. Controlled deployments allow companies to maintain data exclusivity. A hardware sale would put that data generation capability outside the vendor's control.
  1. Capital structure. Most commercial humanoid companies are not yet profitable on a per-unit hardware sale basis. The RaaS model spreads the hardware cost over the contract lifetime, improving unit economics and enabling lower upfront cost for operators.

The Wright's Law Projection: The Consumer Humanoid Market

If Wright's Law holds for humanoid robots — and the BOM trajectory from $130,000 to $50,000 in three years suggests it is already operating — the consumer humanoid market is a question of when, not whether.

The projection:

The critical gating factor is not hardware cost — Wright's Law will handle hardware cost. The gating factor is AI capability: a $5,000 humanoid that cannot reliably perform useful household tasks is a novelty, not a product. The Physical AI development race — the training data, foundation models, and fine-tuning infrastructure that gives robots general-purpose dexterity — is the determinant of the consumer market timeline.

For a deeper look, see What Is Physical AI? and Humanoid Robot Complexity Explained.

The Cricket's Take

> The Unitree G1 costs $16,000. The Agility Digit costs $10/hour to operate. These are not the same product. The G1 is a research platform that happens to be shaped like a human. The Digit is a warehouse worker that happens to be shaped like a human. The conflation of these two things in most humanoid robot coverage is why the market is so poorly understood by investors, journalists, and procurement managers alike.

The 88/100 Jobs Index score for warehouse pickers is not a projection about a future that might happen. It is a score that reflects commercial deployments that are operational today at GXO and Amazon. The crossover economics work at $10/hour against a $27–34/hour fully-loaded human associate, especially at the utilisation rates that robots can sustain and humans cannot. What is not operational today is the generalist humanoid that performs every warehouse task — that is a 2028–2030 story. What is operational is the humanoid that performs the specific tasks it was trained and validated for.

Frequently Asked Questions

How much does a humanoid robot cost in 2026?

Humanoid robot pricing in 2026 spans two fundamentally different market tiers. Research and accessible platforms: Unitree G1 at $16,000 (the most accessible humanoid ever produced), Unitree H1 at $90,000. Commercial deployment platforms: Agility Digit at approximately $10/hour on RaaS, Figure F.02 at enterprise contract pricing (not publicly disclosed), Apptronik Apollo and 1X NEO Beta at commercial programme pricing. Boston Dynamics Atlas is not commercially available for general purchase. The median hardware price of humanoid platforms with confirmed pricing sits above $80,000.

Can I buy an Agility Digit or Figure robot?

No — not as a general market purchase. Agility Digit, Figure F.02, Apptronik Apollo, Boston Dynamics Atlas, and 1X NEO Beta are deployed exclusively through controlled commercial programmes, enterprise contracts, or research partner agreements. These platforms are not available for individual or general corporate purchase. Commercial deployment enquiries go through each company's enterprise sales process. The Unitree G1 ($16,000) and H1 ($90,000) are available for direct purchase.

What does the $10/hour Digit pricing mean compared to human labour?

A warehouse associate costs approximately $27–34/hour fully loaded (wages, benefits, workers' compensation, supervision, and turnover costs at the industry's 60%+ annual attrition rate). At $10/hour, Digit does not generate a dramatic saving at standard single-shift hours. The advantage is at sustained utilisation: a Digit running 20 hours/day, 365 days/year, generates approximately 7,300 hours of productive labour at a cost of $73,000 — equivalent to approximately 3.5 human full-time equivalents at standard hours and standard cost.

Why are most commercial humanoid robots not for sale?

Commercial humanoid platforms are offered as Robot-as-a-Service (RaaS) rather than hardware sales for four reasons: deployment quality control (maintaining performance standards requires vendor involvement), the service is the actual product (performance guarantees, not hardware), data accumulation strategy (operational data is proprietary and valuable), and capital structure (spreading hardware cost over contract lifetime improves unit economics for both vendor and operator). For a full explanation, see What Is RaaS?

What is the Jobs Index score for warehouse workers?

Geppetto's Jobs Index rates warehouse picker/packer roles at 88/100 on the automation risk scale — one of the highest scores in the index. The score reflects confirmed commercial deployment of humanoid and AMR platforms (Agility Digit at GXO, Amazon Sequoia, Locus Robotics) performing warehouse picking and packing tasks at scale. The 88 is not a forward projection — it reflects the current state of commercial deployment.

How fast are humanoid robot costs declining?

According to ARK Investment Management's 2026 research, US humanoid bill of materials costs declined from approximately $130,000 in 2023 to approximately $50,000 in 2026 — a 62% reduction in three years. China-manufactured humanoids (Unitree) are already at the frontier of this cost curve, with the G1's $16,000 retail price reflecting a BOM now well below $10,000. ARK's Wright's Law projection suggests a sub-$10,000 BOM within 8–12 years of mass production (annual volumes of 100,000+ units).

Further Reading