XPENG IRON Review: Specs, AI, Price & Availability

XPENG IRON: Quick Buyer Verdict

The XPENG IRON should currently be evaluated as a pre-launch general-purpose humanoid platform entering industrialisation, not as a finished robot available for ordinary procurement.

That distinction matters. XPENG has moved beyond a stage-only concept: in September 2026 it commissioned dedicated humanoid production lines, said more than 80% of their core processes are automated and showed an IRON autonomously walking off the line after assembly. However, XPENG still describes mass production as a year-end target and external customer deliveries as a 2027 milestone.

Pros

• Dedicated humanoid production line is already commissioned.
• More than 80% automation across the production line’s core processes according to XPENG.
• Highly human-like body architecture and gait.
• Latest specification states 76 degrees of freedom.
• 21 degrees of freedom in each hand.
• Three proprietary Turing AI chips with up to 2,250 TOPS.
• On-device Physical AI model execution reduces dependence on remote inference.
• XPENG controls major parts of the stack, including chips, controllers, motion modules and hands.
• Commercial deployment roadmap is clearer than for many experimental humanoids.
• XPENG already operates global manufacturing, sales and service infrastructure through its automotive business.

Cons

• No official retail price.
• No normal public ordering process in 2026.
• External customer deliveries are not planned until 2027.
• XPENG has not published current official height, weight, payload or battery-runtime figures in its latest production announcement.
• Earlier and current XPENG specifications conflict on DoF and computing power.
• No public standard grasp-force or manipulation-payload specification.
• No detailed public industrial cycle-time or task-success data.
• No mature public SDK documentation comparable with established research humanoids.
• No public IP rating or comprehensive operating-environment specification.
• Many autonomy claims currently come from XPENG rather than independent customer deployments.

Our recommendation: do not treat IRON as an immediately purchasable alternative to an available research robot. Treat it as one of the strongest humanoid programmes to monitor for 2027 procurement. Enterprise buyers should define the intended task now, request commercial and technical information when XPENG opens external sales, and compare the final production configuration against robots that already have measurable customer deployments.

If you are evaluating the category rather than one model, compare current humanoid robots, the best humanoid robots and our guide to humanoid robot pricing.

How Much Does the XPENG IRON Cost?

XPENG has not published an official price for IRON.

As of 12 September 2026, there is no standard MSRP, online checkout process or publicly documented commercial configuration that allows a buyer to calculate the complete cost of an IRON deployment.

This is consistent with the current rollout. XPENG plans mass production by the end of 2026, followed by initial deployments in its own stores and campuses. Official launch and large-scale external deliveries in China and overseas are planned for 2027.

Do not use unofficial IRON price estimates

A recurring problem with pre-launch humanoid robots is that speculative price figures get repeated until they look official.

Until XPENG publishes an MSRP, quote, lease structure or commercial contract, the accurate answer is simple: the price is not public.

That also means IRON cannot yet be compared fairly with an orderable robot using headline price alone.

A real procurement comparison will eventually need:

• Robot hardware price.
• AI software or model subscription costs.
• Maintenance and support.
• Battery replacement.
• Charging infrastructure.
• Training and integration.
• Replacement parts.
• Connectivity and fleet management.
• Custom application development.
• Warranty and service response.

XPENG has indicated that recurring revenue from future AI-model upgrades could become part of the lifetime economics of each IRON, suggesting buyers should not assume a one-time hardware purchase will represent the complete ownership cost.

What Is the XPENG IRON?

The XPENG IRON is a full-size humanoid robot developed by Chinese Physical AI and electric-vehicle company XPENG.

The project has been under development for several years and shares technology with XPENG’s broader AI stack, including its proprietary Turing AI chips and physical-world foundation models. XPENG’s objective is not merely to build a walking machine but a general-purpose humanoid capable of perceiving, reasoning, moving and interacting in environments originally designed for people.

Its most distinctive design decision is what XPENG calls extreme anthropomorphism.

Rather than exposing an industrial frame, IRON uses human-like body proportions, an articulated spine and waist, compact actuators, highly articulated hands and a fully enclosed exterior. XPENG says the design allows human behavioural data to transfer more naturally to the robot while making it better suited to spaces, objects and workflows built around the human body.

What IRON is

• A general-purpose humanoid platform under active commercial development.
• A physical embodiment for XPENG’s Physical AI models.
• A robot designed for human-scale environments.
• A platform prioritising natural walking and human-like movement.
• A potential retail, service and future industrial robot.
• A vertically integrated product using major XPENG-developed components.
• A robot transitioning from prototypes into production-line manufacturing.

What IRON is not

• It is not currently a normally orderable commercial robot.
• It is not proven to perform arbitrary household or industrial tasks.
• It does not have a published customer price.
• It does not have a complete public production specification sheet.
• It does not yet have extensive published third-party productivity evidence.
• It is not equivalent to every prototype or specification XPENG has shown since 2024.
• It is not proof that general-purpose humanoid autonomy has been solved.

XPENG IRON 2024 vs 2025 vs 2026

Understanding IRON requires separating three different stages of the programme.

XPENG has changed the robot substantially since the original 2024 unveiling, and mixing specifications from different generations produces misleading comparisons.

The 2024 robot was described by XPENG as having more than 60 joints and 200 degrees of freedom and was primarily focused on internal factory and retail applications.

In November 2025, XPENG presented a dramatically redesigned Next-Gen IRON with a human-like spine, bionic muscles, flexible skin, 82 DoF and 22 DoF in each hand. That announcement also described an all-solid-state battery and 3,000 TOPS of effective compute.

The September 2026 production-line announcement gives different numbers: 76 DoF, 21 DoF per hand and three Turing chips delivering up to 2,250 TOPS.

For a buyer, the newest production-era figures should take priority until XPENG publishes a complete final commercial specification sheet.

XPENG IRON Specifications

The following table separates what XPENG currently publishes from specifications that remain undisclosed.

XPENG’s latest production announcement supports the 76-DoF, 21-DoF-hand and 2,250-TOPS figures.

The missing specifications matter

IRON looks technically advanced, but procurement requires less glamorous numbers.

A buyer evaluating an inspection, retail or material-handling task eventually needs:

• Payload.
• Reach.
• Gripping force.
• Repeatability.
• Walking speed.
• Minimum aisle width.
• Step and stair capability.
• Runtime.
• Recharge time.
• Duty cycle.
• Operating temperature.
• Ingress protection.
• Fall behaviour.
• Stopping distance.
• Service intervals.

Until those are published or provided in a commercial quotation, IRON cannot be evaluated with the same precision as an established industrial robot.

82 DoF vs 76 DoF: Which XPENG IRON Specifications Are Current?

Use 76 DoF and 21 DoF per hand as the current figures.

This is one of the most important details in the XPENG IRON story.

XPENG’s November 2025 AI Day described Next-Gen IRON with 82 degrees of freedom, 22 DoF in each hand and 3,000 TOPS.

XPENG’s September 8, 2026 production-line announcement instead describes the robot with 76 DoF, 21 DoF per hand and 2,250 TOPS.

XPENG has not publicly explained every engineering change between these specifications.

Possible reasons could include production simplification, revised DoF counting, cost, reliability, packaging or hardware optimisation, but buyers should not speculate.

The important procurement rule is simpler:

Do not combine the best specification from every IRON prototype into one imaginary production robot.

Request the exact configuration being sold.

Walking, Balance and Human-Like Movement

Movement is currently IRON’s most visibly differentiated capability.

XPENG deliberately designed the robot to move less like a conventional industrial machine and more like a person.

The company says engineers used human gait data, motion remapping, reinforcement learning and simulation to develop IRON’s distinctive walking behaviour. Its revised shoulders, waist and articulated spine allow bending, twisting, arm swing and body movement that make the gait appear unusually natural.

Why does IRON walk like a human?

XPENG’s approach begins with physical structure rather than adding human-like motion to a conventional robot afterwards.

The robot uses:

• A human-like spine and waist architecture.
• Compact joint actuators.
• Human-proportioned limbs.
• Coordinated upper- and lower-body motion.
• Motion captured from human models.
• Reinforcement-learning-based gait development.

This lets IRON use shoulder movement, waist rotation and arm inertia while walking instead of keeping the torso rigid.

Does human-like walking make IRON more useful?

Potentially.

Natural body mechanics can matter when the robot must:

• Navigate environments designed around humans.
• Interact closely with customers.
• Reach shelves, displays or controls.
• Learn motion from human demonstrations.
• Transfer human behavioural data into robot training.
• Perform tasks requiring whole-body coordination.

However, natural-looking gait is not the same as robust mobility.

For commercial operation, buyers still need evidence covering:

• Hours between falls.
• Floor transitions.
• Ramps.
• Stairs.
• Obstacle recovery.
• Slippery surfaces.
• Crowd navigation.
• Emergency stopping.
• Recovery after contact.

XPENG’s public materials do not yet provide a complete benchmark for those conditions.

XPENG IRON Hands and Manipulation

IRON’s hands are another major part of its human-like design.

XPENG’s latest published configuration states 21 degrees of freedom in each hand. Earlier Next-Gen IRON material described 22-DoF hands built around extremely compact harmonic joints and approximately human-scale proportions.

High articulation theoretically helps with:

• Different grasp shapes.
• Finger positioning.
• Two-handed manipulation.
• Human-tool compatibility.
• Opening drawers and doors.
• Handling consumer objects.
• Retail interaction.
• Human-to-robot motion transfer.

What we do not know yet

XPENG has not published the complete manipulation specification a serious industrial buyer would normally require.

Important missing figures include:

• Maximum hand payload.
• Arm payload.
• Pinch force.
• Grip force.
• Finger repeatability.
• Positioning accuracy.
• Maximum reach.
• End-effector speed.
• Continuous manipulation duty cycle.
• Expected hand service life.

A hand can have many joints and still struggle with reliable production manipulation.

The commercial test should therefore be task based.

Buyer rule: do not ask whether IRON has dexterous hands. Ask whether the exact production robot can pick up your object, move it through the required path, place it to the required tolerance and repeat that process reliably for an entire shift.

AI, Turing Chips and Autonomous Capabilities

IRON is important because XPENG is attempting to combine humanoid hardware with an AI stack already developed across vehicles, Robotaxis and other Physical AI systems.

The latest production-era IRON uses three XPENG Turing AI chips delivering up to 2,250 TOPS. XPENG says this allows its Physical AI foundation model to run locally on the robot, reducing inference latency and allowing complex tasks to be performed without relying on remote operation.

That is an unusually large amount of advertised onboard compute for a humanoid.

VLA, VLM and VLT

XPENG has previously described IRON’s intelligence architecture as combining:

• VLA: vision-language-action capabilities linking perception and physical actions.
• VLM: vision-language understanding for interpreting scenes and communication.
• VLT: XPENG’s vision-language-task model intended to support task reasoning and autonomous decision-making.

XPENG described this architecture in its 2025 Next-Gen IRON materials as a high-level brain and cerebellum system intended to support environmental inference and autonomous decisions.

Does IRON really work autonomously?

XPENG says the production-era platform can run its foundation model onboard and autonomously perform complex tasks without remote operation.

That is significant, but it should still be treated as a manufacturer capability claim rather than proof of general-purpose autonomy.

For a buyer, useful autonomy has to be measured through numbers such as:

• Task success rate.
• Interventions per operating hour.
• Average recovery time.
• Navigation failures.
• Manipulation failures.
• Performance across unfamiliar environments.
• Time required to learn a new task.
• Human supervision requirements.

Those metrics are not yet publicly established for IRON.

On-device AI is still important

Even without assuming general autonomy, running large models onboard can provide several practical advantages:

• Lower control latency.
• Reduced reliance on cloud connectivity.
• More predictable operation in poor-network conditions.
• Potentially better privacy.
• Faster perception-action loops.
• Greater independence from remote operators.

XPENG’s broader strategy is to reuse technology and data across vehicles, Robotaxis and robots rather than developing IRON’s intelligence in isolation.

XPENG IRON SDK and Developer Access

XPENG announced in 2025 that it intended to open the IRON SDK and develop a global humanoid application ecosystem.

That is strategically important, but buyers should distinguish an announced open ecosystem from a mature developer platform.

The current public XPENG materials reviewed for this article do not provide the same depth of robot-specific developer documentation, repositories, simulation environments and API specifications available for established research platforms.

What developers should ask XPENG

Before treating IRON as an open development platform, confirm:

• Whether the SDK is available to external customers.
• Which operating systems are supported.
• Whether ROS or ROS 2 interfaces are officially supported.
• Access to joint-level control.
• Access to high-level navigation commands.
• Camera and sensor APIs.
• Hand-control APIs.
• Simulation models.
• Teleoperation interfaces.
• Onboard model deployment.
• Third-party application permissions.
• Software licensing.
• OTA update policy.
• Safety restrictions on low-level control.

For research teams that need deep control immediately, an orderable developer-focused robot may remain easier to work with in 2026.

Battery Life, Charging and Operating Endurance

Battery technology is one area where IRON’s public specification requires particular caution.

At XPENG AI Day 2025, the company described Next-Gen IRON as using an all-solid-state battery, presenting it as a route to lower weight, higher energy density and improved safety.

However, XPENG’s September 2026 production-line announcement does not provide battery capacity, chemistry, runtime or recharge time.

Therefore, this review does not assume that every production IRON delivered in 2027 will use exactly the battery configuration announced in November 2025.

What buyers need to know

Before deployment, request:

• Battery chemistry.
• Nominal and usable energy capacity.
• Typical walking runtime.
• Runtime during manipulation.
• Idle runtime.
• Recharge time.
• Whether autonomous charging is supported.
• Whether batteries can be swapped.
• Cycle-life specification.
• Battery replacement cost.
• Thermal operating range.
• Low-battery safety behaviour.

Runtime is particularly important for IRON’s planned retail and service applications.

A robot that can interact intelligently but requires long charging periods after short operating windows may need multiple units or a different charging strategy to cover a commercial day.

XPENG IRON Safety and Operating Limitations

XPENG places unusual emphasis on designing IRON to coexist with people.

Its current robot uses a fully enclosed flexible lattice structure intended to balance appearance and safety, while the earlier Next-Gen platform used a fully covered flexible exterior around the mechanical structure.

This is sensible for a robot intended to work in retail and service environments.

However, a soft-looking exterior does not automatically make a humanoid safe.

Public safety information is still limited

A commercial buyer should request data covering:

• Emergency-stop behaviour.
• Maximum contact force.
• Collision detection.
• Pinch points.
• Finger-force limits.
• Fall detection.
• Fall direction and exclusion zones.
• Stopping distance.
• Maximum safe walking speed around people.
• Battery safety.
• Functional-safety architecture.
• Relevant certifications and standards.

These details are more important than whether the exterior feels soft.

Privacy and onboard processing

XPENG previously described a privacy principle under which personal data should remain on the robot. Its current architecture also emphasises onboard AI inference and enhanced data security rather than depending entirely on cloud processing.

That direction is valuable for customer-facing deployments, but enterprise buyers should still confirm:

• What video and audio IRON records.
• Where recordings are stored.
• Whether data leaves the robot.
• How long information is retained.
• Who can access it.
• Whether recordings are used for model training.
• Whether remote access can be disabled.
• How security updates are authenticated.

XPENG IRON Production and Commercial Readiness

This is where IRON became significantly more interesting in September 2026.

On 8 September, XPENG announced that its dedicated humanoid manufacturing facility had entered operation and that an IRON completed production and autonomously walked off the new line. XPENG says more than 80% of the line’s core processes are automated.

That is materially different from showing another hand-built prototype.

What the production line proves

It provides evidence that XPENG is investing in:

• Repeatable assembly processes.
• Production tooling.
• Automated manufacturing.
• Quality control.
• Component integration.
• Future capacity expansion.

XPENG is transferring manufacturing practices from its electric-vehicle business into humanoid production.

What the production line does not prove

It does not yet establish:

• Sustained monthly production volume.
• Manufacturing yield.
• Customer delivery volume.
• Robot reliability.
• Field service costs.
• External customer satisfaction.
• Commercial task productivity.

XPENG itself still describes mass production as a future milestone targeted for the end of 2026. Customer launch and deliveries follow in 2027.

How many XPENG IRON robots could be produced?

XPENG said in its Q2 2026 results that monthly production capacity could be rapidly increased to several thousand units during 2027 in response to market demand.

That is a capacity roadmap, not a current shipment figure.

The difference matters.

A factory capable of producing thousands of robots is not the same as thousands of robots operating successfully for paying customers.

What Real-World XPENG IRON Evidence Shows

IRON has moved through several levels of evidence, but they should not be treated as equivalent.

During the 2026 Spring Festival, XPENG reported deploying IRON at the Beijing Auto Museum, where it interacted with visiting children. This provides evidence of public-facing use beyond an internal engineering laboratory.

XPENG has also named guided tours, shopping assistance and Baosteel industrial inspection as target applications.

The strongest new manufacturing evidence came in September 2026 when an IRON completed XPENG’s newly commissioned production process and walked off the line.

What the evidence supports

• IRON is a real physical humanoid programme.
• XPENG has developed sophisticated bipedal motion.
• The robot can operate in controlled public-facing demonstrations.
• XPENG has moved into production-line manufacturing.
• There is a defined commercial rollout strategy.

What buyers still need

• Independent operating-hour data.
• Task-success percentages.
• Intervention rates.
• Battery endurance.
• Manipulation benchmarks.
• Industrial cycle times.
• Mean time between failures.
• External customer case studies.
• Total cost of ownership.

The evidence is therefore stronger than a concept robot, but still weaker than a mature commercial automation product.

Best Uses for the XPENG IRON

1. Retail sales assistance

This is IRON’s clearest near-term commercial use case.

XPENG has stated that the robot will begin working as an AI-powered retail assistant in its Chinese stores, with overseas rollout planned afterwards.

Retail uses can combine:

• Customer greeting.
• Product explanation.
• Questions and answers.
• Navigation around a showroom.
• Brand engagement.
• Lead capture.
• Physical demonstrations.

These tasks are comparatively forgiving because human staff remain nearby and physical manipulation requirements can be limited.

2. Guided tours and visitor experiences

Museums, visitor centres, corporate campuses and exhibitions fit IRON’s strengths.

Its appearance, speech, mobility and human-like movement create an experience that a screen or wheeled kiosk cannot reproduce.

XPENG’s Beijing Auto Museum deployment provides an early example of this direction.

3. Corporate campuses

XPENG says campuses will form part of the initial deployment phase.

Possible applications include:

• Reception.
• Visitor guidance.
• Internal demonstrations.
• Wayfinding.
• Employee interaction.
• Physical-AI data collection.

4. Industrial inspection research

XPENG has identified Baosteel as a partner for exploring complex industrial inspection.

A humanoid could theoretically navigate human workspaces, inspect gauges, approach existing equipment and interact with controls without redesigning the complete facility.

However, this use case needs stronger evidence covering environmental protection, navigation reliability, runtime and safety before broad deployment.

5. Physical-AI development

IRON can become an important platform for XPENG’s own embodied-AI development because production-scale fleets create more real-world robot data.

This may ultimately be one of its biggest advantages.

More robots create more interaction data; more data can improve models; improved models can enable additional applications.

6. Brand activation and exhibitions

IRON’s extremely human-like gait makes it unusually effective as a technology showcase.

This is not the most economically important long-term use, but it can provide immediate commercial value for:

• Automotive showrooms.
• Trade shows.
• Technology events.
• Luxury retail.
• Corporate exhibitions.

7. Future service work

XPENG expects external customers in retail and service sectors from 2027.

The longer-term opportunity includes tasks that combine:

• Walking.
• Conversation.
• Scene understanding.
• Object interaction.
• Human-compatible reach.

But each proposed service task still needs individual validation.

When the XPENG IRON Is Not the Right Robot

IRON should not automatically be selected because humanoids are exciting.

Reject or delay IRON when:

• You need a robot today: normal external deliveries are planned for 2027.
• You need a fixed budget: there is no published price.
• You require known payload: XPENG has not published a current production payload specification.
• You require all-day runtime: public production battery-endurance figures are unavailable.
• You need mature industrial KPIs: cycle-time and reliability evidence remain limited.
• You need an open research platform immediately: public developer resources are still developing.
• The task is repetitive and fixed: a conventional industrial arm may be cheaper, faster and easier to validate.
• The task is material transport: an AMR may offer better payload and endurance.
• The task is outdoor inspection: a weather-rated quadruped may be more appropriate.
• You need safety-certified close collaboration: require written evidence before assuming IRON meets the application.

The humanoid form makes sense when human geometry actually solves a constraint.

If wheels, a fixed arm or a simpler service robot can perform the task reliably, the simpler machine may still be the better investment.

XPENG IRON vs Figure 03, Tesla Optimus, Unitree G1 and UBTECH Walker S2

IRON sits in a difficult comparison group because these robots target different buyers.

Figure 03 currently provides stronger field evidence. Figure says its third-generation robot has entered BMW logistics work, while its published specification includes a 61 kg body, 20 kg payload and five-hour runtime.

Tesla Optimus remains another major automotive-backed programme, although Tesla’s latest filings still describe production-line installation and initial internal builds rather than external commercial sales.

The Unitree G1 serves a different market: it is an orderable development platform rather than a vertically integrated future service worker. Its commercial availability makes it much easier for laboratories that need hardware now.

UBTECH Walker S2 is more industrially focused. UBTECH describes a 52-DoF body, up to 15 kg handling and an autonomous three-minute battery-swap system designed to support continuous operation.

Which should you choose?

• Track XPENG IRON if you want a future human-like retail or service humanoid with serious manufacturing backing.
• Choose Figure 03 when current real-world AI manipulation evidence matters most.
• Track Tesla Optimus when long-term manufacturing scale and Tesla’s AI ecosystem are the priority.
• Choose Unitree G1 EDU when you need a programmable humanoid development platform now.
• Choose Walker S2 when the requirement is industrial handling and long operating availability.

Use the Anton Robots comparison tool to compare current robots by application and specification.

Is the XPENG IRON Worth It?

It is too early to say whether XPENG IRON is worth buying because XPENG has not yet published its price or opened normal customer sales.

What can already be judged is the programme.

And the programme is unusually credible.

XPENG has:

• Built several generations of humanoid hardware.
• Developed proprietary AI chips.
• Built a shared Physical AI model stack.
• Transferred automotive manufacturing capability into robotics.
• Commissioned a dedicated humanoid production line.
• Raised more than US$900 million for its robotics business.
• Defined internal and external deployment phases.

XPENG’s robotics funding round valued the business at more than US$6.3 billion and is intended partly to accelerate humanoid mass production and AI development.

That does not guarantee commercial success, but it significantly reduces the risk that IRON is merely a temporary demonstration project.

Where IRON could create value

IRON becomes compelling if XPENG can combine:

• Human-compatible movement.
• Useful hand manipulation.
• Reliable autonomous navigation.
• Natural customer interaction.
• Long-enough operating endurance.
• Mass-manufacturing economics.

The largest uncertainty is not whether XPENG can build an impressive humanoid.

It is whether IRON can perform enough useful work, reliably enough and cheaply enough, to beat specialised robots or human labour in specific applications.

The value test

Before considering IRON, complete this sentence:

We need a humanoid because the task requires ________, and a wheeled robot, fixed robot arm or existing service robot cannot solve that requirement efficiently.

Good answers might include human-scale navigation, moving between existing workstations, manipulating equipment designed for hands or combining customer interaction with physical tasks.

“IRON looks human” is not enough.

XPENG IRON Buying Checklist

1. Define the task. Write one exact workflow the robot must complete.
2. Confirm commercial availability. Request the launch date for your country.
3. Request the final production specification. Do not rely on 2025 prototype figures.
4. Confirm the price. Separate hardware, software, support and recurring AI costs.
5. Confirm robot dimensions and weight. Check doors, lifts, floors and transport.
6. Define payload. Request arm and hand limits for the exact working posture.
7. Test manipulation. Use the real objects the robot must handle.
8. Measure autonomy. Record task success and human interventions.
9. Validate navigation. Test crowds, obstacles and changing environments.
10. Confirm runtime. Measure productive operating time rather than headline battery duration.
11. Review charging. Confirm recharge time and autonomous or manual charging requirements.
12. Review developer access. Obtain SDK, API and software documentation.
13. Review privacy. Define how camera, microphone and interaction data are processed.
14. Review safety. Require emergency-stop, collision and fall documentation.
15. Confirm environmental limits. Obtain temperature, dust and moisture specifications.
16. Confirm warranty. Include actuators, hands, battery, cameras and compute.
17. Confirm repairs. Identify local service capability and spare-parts lead time.
18. Run an acceptance test. Test the exact production robot before deployment.
19. Measure economics. Compare cost per successful task with alternative automation.

Pro tip: because IRON has already changed specifications between its 2025 unveiling and 2026 production phase, attach the final configuration sheet directly to the purchase contract. Do not buy based on specifications copied from an older demonstration.

How to Buy the XPENG IRON

You cannot currently buy XPENG IRON through a normal public ordering process.

XPENG’s current roadmap is:

1. Production-line manufacturing in 2026.
2. Mass production targeted by the end of 2026.
3. Initial deployment in XPENG stores and campuses.
4. Official launch in 2027.
5. Large-scale deliveries to external customers in China and overseas.

XPENG says those external customers will initially focus on retail and service applications.

What should an interested buyer do now?

Prepare a structured requirement rather than simply asking for an IRON quotation.

Include:

• Country.
• Industry.
• Deployment environment.
• Exact task.
• Objects the robot must manipulate.
• Required operating hours.
• Human interaction requirements.
• Number of robots.
• Target deployment date.
• Required software integration.
• Safety requirements.
• Expected support level.

You can also contact Anton Robots to compare IRON with currently available humanoids or use Find My Robot if the application matters more than a specific brand.

What Is New for XPENG IRON in 2026?

A real production line

The biggest development is manufacturing.

On September 8, 2026, XPENG announced the commissioning of its humanoid production lines and showed an IRON completing the manufacturing process and walking away autonomously. More than 80% of core processes are automated according to XPENG.

This is the clearest sign yet that the programme is moving from experimental hardware toward industrialised production.

The production specification changed

The current robot is described with 76 DoF, 21 DoF per hand and 2,250 TOPS.

That replaces the 82-DoF, 22-DoF-hand and 3,000-TOPS figures associated with the November 2025 unveiling.

More than US$900 million for XPENG Robotics

In August 2026, XPENG announced that its robotics business had raised more than US$900 million at a post-money valuation above US$6.3 billion.

The funding is intended to accelerate Physical AI research, humanoid mass production and commercialisation.

2027 external deliveries are now explicit

XPENG now says IRON will officially launch in 2027 and begin large-scale deliveries to customers in China and overseas, initially focused on retail and service sectors.

Retail deployment has become more specific

XPENG Australia said IRON is intended to begin operating as an AI-powered retail sales assistant in Chinese XPENG stores from Q1 2027, followed by overseas rollout later in the year.

That gives the programme a much more concrete first commercial application than it had at the beginning of 2026.

XPENG IRON FAQ

What is XPENG IRON?

XPENG IRON is a general-purpose humanoid robot developed by XPENG as part of its Physical AI strategy. It combines bipedal locomotion, highly articulated hands and proprietary onboard AI computing.

Is XPENG IRON real?

Yes. Physical IRON robots have appeared at public events and real venues, and XPENG commissioned a dedicated production line in September 2026.

Can you buy XPENG IRON?

Not through a normal public ordering process in 2026. XPENG plans its official market launch and external customer deliveries for 2027.

How much does XPENG IRON cost?

XPENG has not published an official price. Treat third-party price estimates as unofficial unless XPENG confirms them.

When will XPENG IRON launch?

XPENG says official launch and large-scale deliveries in China and overseas are planned for 2027.

When will XPENG IRON enter mass production?

XPENG targets mass production by the end of 2026. Its dedicated production line was commissioned in September 2026.

How many degrees of freedom does XPENG IRON have?

The latest September 2026 XPENG specification states 76 DoF.

Why do some websites say XPENG IRON has 82 DoF?

Because XPENG’s November 2025 Next-Gen IRON announcement specified 82 DoF. The newer September 2026 production announcement specifies 76, so this review uses 76 as the current figure.

How many degrees of freedom do the hands have?

XPENG’s latest figure is 21 DoF per hand. The 2025 prototype-era specification stated 22.

How powerful is XPENG IRON’s AI computer?

XPENG currently states that three Turing AI chips provide up to 2,250 TOPS of effective computing power.

Why do some reports say 3,000 TOPS?

XPENG announced 3,000 TOPS for Next-Gen IRON in November 2025. Its September 2026 production-era announcement states 2,250 TOPS.

Does XPENG IRON use AI?

Yes. IRON is designed around XPENG’s Physical AI foundation-model stack, and XPENG says models run locally using its Turing chips.

Is XPENG IRON autonomous?

XPENG says the robot can use onboard models to perform complex tasks without remote operation. Public independent performance data are not yet sufficient to treat IRON as a proven unrestricted general-purpose autonomous worker.

Is IRON remotely controlled?

XPENG says its current architecture supports autonomous onboard task execution. Individual demonstrations should still be evaluated separately because public performance does not prove that every action shown is autonomous.

Can XPENG IRON talk?

Yes. Human interaction and dialogue are central parts of XPENG’s intended customer-facing use cases.

Can XPENG IRON work in shops?

That is its clearest planned first commercial application. XPENG intends to deploy IRON as a retail assistant in its own stores before broader external sales.

Can XPENG IRON work in a factory?

Potentially. XPENG has explored factory and industrial applications and announced a Baosteel inspection partnership, but public productivity evidence remains limited.

Can XPENG IRON manipulate objects?

IRON has highly articulated hands intended for manipulation, but XPENG has not published a complete production payload, grip-force or manipulation-success specification.

How much can XPENG IRON carry?

XPENG has not published a current official production payload figure.

How fast can XPENG IRON walk?

XPENG’s latest production announcement does not provide a standard maximum walking-speed specification.

How long does the XPENG IRON battery last?

XPENG has not published a current official runtime figure.

Does XPENG IRON use a solid-state battery?

XPENG announced an all-solid-state battery for Next-Gen IRON in November 2025. The September 2026 production announcement does not restate battery chemistry, so buyers should confirm the final commercial configuration.

How tall is XPENG IRON?

The latest official production announcement reviewed here does not provide a current height specification. Avoid treating third-party figures as final production specifications.

How much does XPENG IRON weigh?

XPENG has not published a current production weight in the latest specifications reviewed here.

Does XPENG IRON have an SDK?

XPENG has announced plans to open its SDK and build a developer ecosystem. Buyers should confirm current availability, interfaces and licensing before planning a development programme.

Does XPENG IRON support ROS?

XPENG has not published enough current robot-specific developer documentation for this review to confirm a standard production ROS integration.

Is XPENG IRON safe around people?

XPENG is designing IRON for human environments and uses a flexible enclosed exterior, but buyers still need formal safety specifications, risk assessment and application-specific validation.

Where has XPENG IRON been deployed?

XPENG has reported IRON operating at the Beijing Auto Museum and has used robots internally. Initial commercial deployments are planned for XPENG stores and campuses.

Will XPENG IRON be sold outside China?

Yes, according to XPENG’s current roadmap. Overseas external deliveries are planned from 2027.

Will XPENG IRON come to Australia?

XPENG has already displayed IRON in Australia and has stated that overseas rollout is planned after its initial Chinese retail deployment. A specific Australian customer-sales date and price have not yet been published.

Is XPENG IRON better than Tesla Optimus?

It is too early to make a definitive comparison. IRON currently has a commissioned production line and detailed published compute figures, while Optimus benefits from Tesla’s AI and manufacturing ecosystem. Neither has a normal public commercial purchase route today.

Is XPENG IRON better than Figure 03?

Figure 03 currently has stronger published real-world deployment evidence, including work at BMW. IRON differentiates itself through human-like movement, high onboard compute and XPENG’s integrated Physical AI strategy.

Is XPENG IRON better than Unitree G1?

They target different buyers. Unitree G1 EDU can be purchased as a development platform today. IRON is aimed at future integrated commercial service applications and is not yet generally available.

Is XPENG IRON worth buying?

There is not enough commercial information to answer yet. Price, production configuration, support, runtime and field reliability must be known before calculating value.

Final Verdict: Should You Buy the XPENG IRON?

XPENG IRON is one of the humanoid robots most worth watching for 2027, but it is not yet a normal buying decision in 2026.

The important story is no longer simply the robot’s remarkably human-like walk.

XPENG has now connected several pieces that many humanoid programmes still treat separately: proprietary AI chips, Physical AI models, human-like mechanical design, autonomous onboard inference, global automotive manufacturing experience and a dedicated humanoid production line.

That makes IRON significantly more credible than a stage prototype.

But buyers should resist the opposite mistake.

IRON does not yet have a public price, normal order path, complete production specification, published runtime, payload, mature developer documentation or a large body of external customer performance data.

The September 2026 production milestone proves that XPENG is industrialising the robot. It does not yet prove that IRON can deliver reliable economic value across general-purpose work.

For now, the strongest buyer position is to shortlist and monitor IRON rather than budget around assumptions.

When commercial quotations open, ask XPENG for the exact production configuration, then validate one defined task against measurable acceptance criteria: autonomy, task success, runtime, manipulation, safety, serviceability and total cost.

If XPENG can combine the manufacturing capability it has already demonstrated with reliable autonomous work, IRON could become one of the most important commercial humanoid platforms of the next several years.

Until then, evaluate what exists—not what the humanoid category promises.

Read Anton Robots’ coverage of XPENG IRON’s production milestone, compare current humanoid robots, or contact Anton Robots to evaluate available alternatives.