Figure 03 Robot

How much does a Figure 03 cost?

Figure has never published a price, for anyone. Not a unit price, not a lease rate, not an hourly rate, not a pilot fee. There is no configurator, no preorder and no self-service channel of any kind.

The only consumer-facing number that has ever come from Figure's leadership is a sketch. On a factory tour published in May 2026, CEO Brett Adcock suggested humanoids could enter homes on an automotive-style lease of roughly $400 to $600 a month, and mentioned Figure 03 at something like $600 a month. That is a forward-looking model, not an offer, and no start date was attached to it.

On the enterprise side there are only third-party estimates, and the weak ones are circulating hard. A figure of around $1,000 per robot per month for robotics-as-a-service appears on marketing blogs with no sourcing at all. Estimates of $30,000-$50,000 unit cost for Figure 02 circulate similarly. None of these come from Figure.

Here is our own estimate, with the reasoning shown. Figure does not sell units - it deploys them into partner facilities with Figure engineers embedded on site, which is how both the BMW and Catalyst programmes are structured. That means the dominant cost in a deployment is integration and service labour, not hardware. Against third-party hardware anchors near $50,000 a unit, and robotics-as-a-service anchors in the low-to-mid five figures per robot per year, a realistic first-year enterprise pilot of two to five robots at one site lands somewhere between $150,000 and $500,000, or roughly $20,000-$60,000 per robot-year all-in.

We are labelling that clearly as an estimate. The true figure is unknown because every Figure deal to date is a bespoke, undisclosed partnership, and neither BMW nor Catalyst Brands has disclosed a contract value.

Can you buy or preorder a Figure 03?

No — and this is worth being emphatic about, because the search demand says people believe otherwise.

There is no preorder, no waitlist, no reservation page, no deposit and no price, as of 30 August 2026. Figure's access model is a closed partner programme: robots go into partner facilities alongside Figure engineers. Two commercial customers are public - BMW and Catalyst Brands - and that is the entire list.

The most likely source of the confusion is 1X, which does take real consumer preorders for NEO at $20,000 outright or $499 a month with a $200 deposit. That $20,000 figure has migrated onto Figure in a lot of coverage, and onto Tesla Optimus, where it is also wrong. If a site is taking your money for a Figure 03, it is not authorised by Figure.

There is also a stock angle worth answering, because it appears in the People Also Ask box on nearly every Figure search. Figure AI is private. On 8 July 2026 the company published a notice stating that unauthorised secondary sales of its stock are void, naming several secondary marketplaces by name, and extending the restriction upstream to special-purpose vehicle and fund interests, with board approval required for any transfer. So the answer to "can I buy Figure AI stock" is effectively no, and brokered offers you encounter should be treated as unauthorised.

Is Figure 03 really working at BMW?

Yes - and the honest version of this story has two chapters, one of which is not flattering.

Chapter one, Figure 02. The partnership was announced in January 2024, with Figure 02 working in the body shop inserting sheet-metal parts for welding. In November 2025 Figure published deployment numbers: roughly ten to eleven months of operation, about 1,250 operating hours, more than 90,000 components moved, contributing to production of over 30,000 BMW X3 vehicles. Those figures were later corroborated in BMW's own material.

But in April 2025, Fortune reported that Figure had materially overstated the scale. A BMW spokesperson said only one Figure robot was operating at any given time, and that until around March 2025 it ran only outside production hours on a single limited pick-and-place task - against Figure's earlier public framing of a fleet performing end-to-end operations. Adcock publicly called the article "downright lies" and said Figure's counsel would aggressively pursue remedies including defamation claims. No defamation suit appears ever to have been filed. Doing the arithmetic on BMW's own numbers - 1,250 hours over about eleven months - gives roughly four hours a working day, which is consistent with a modest single-robot operation rather than a fleet.

Chapter two, Figure 03, is stronger and it is documented by BMW rather than by Figure. BMW's press release of 25 June 2026 places Figure 03 at Plant Spartanburg in Assembly Hall 52 on logistics sequencing, framed as a production use case following the successful pilot, and quotes BMW Manufacturing's vice president of production control and logistics describing Spartanburg as the birthplace of humanoid robotics in BMW's operational day-to-day activities. Figure confirmed the arrival on 30 June 2026, and the deployment was still being reported as active in late August 2026.

What nobody has disclosed is how many Figure 03 units are there. Not BMW, not Figure, not any outlet we checked. Given the 2025 history, assume small until someone publishes a number.

How autonomous is Figure 03, really?

This is the area where Figure has the strongest evidence of any Western humanoid maker, and it deserves some credit before the caveats.

Helix is Figure's in-house vision-language-action model, introduced in February 2025 after Figure ended its OpenAI collaboration - Adcock's stated reason being that language models had become a smaller problem than high-rate robot control. Helix 02, announced 27 January 2026, unified vision, touch, proprioception and whole-body control into a single network covering walking, balance and manipulation together, rather than bolting a locomotion controller to a separate manipulation policy. It runs entirely onboard the robot.

The strongest public evidence came in May 2026: a multi-day package-sorting livestream, originally planned for eight hours, that ran to roughly 101,000 packages over 81 continuous hours with, per Adcock, no teleoperation and no human intervention. Figure separately reported robots performing at about 98.5% as well as a human in a sorting comparison. Bloomberg covered the no-teleoperation claim directly. Nothing Tesla or 1X has shown comes close to that as an autonomy demonstration.

Now the caveats, which matter. It was a controlled demonstration in Figure's own facility, not a customer site. Observers noted accuracy problems in the stream, including barcode-orientation errors. And Figure continues to hire teleoperation staff for data collection - teleoperation remains central to how Helix is trained, even where the demos themselves run autonomously.

The home demos - living-room tidying in March 2026, bedroom resetting with two robots on a single learned policy in May 2026 - were also filmed in Figure-controlled house environments. There is no evidence of Helix running autonomously in a customer's home, for the straightforward reason that there is no public customer-home programme.

Is Figure 03 shipping to homes?

No, and the silence around this is itself informative.

Figure announced a home alpha testing programme in February 2025, with Adcock saying it was two years ahead of schedule and would start with Figure 02 in late 2025. Since then, across Figure's entire 2026 newsroom - eight posts between January and August - there has been no home programme update, no count of homes, and no participant disclosure. The home-context material Figure published in 2026 is demonstration footage shot in Figure-controlled environments, not a customer programme.

That leaves the home alpha's status genuinely unverified as of 30 August 2026. Our reading is that it is small and private at best.

There is a structural reason to be sceptical of any near-term home launch, and it is certification rather than software. The only certification Figure has publicly claimed is on the battery, and it is described as in process rather than achieved - dual certification to UN38.3 and UL2271, developed with OSHA-recognised testing laboratories. There is no public evidence of any whole-robot safety certification: no UL 3300, no ISO 13482, no ISO 10218 claim. A humanoid in a private home would realistically need UL 3300 for service robots, ISO 13482 for personal care robots, household electrical safety approval, plus electromagnetic compatibility, functional safety and a cybersecurity assessment.

Figure has published no home-certification timeline. Until a certification listing appears, treat home availability as a research goal rather than a product roadmap.

What is the Gruendel lawsuit about?

It needs stating precisely, because the viral version of it is wrong in a way that actually matters.

Robert Gruendel, a robotics safety engineer previously at Amazon's R&D group, was hired by Figure as head of product safety. On 21 November 2025 he filed suit in the US District Court for the Northern District of California under California whistleblower-protection statutes and for wrongful termination in violation of public policy, with a jury trial demanded.

The allegations, as pleaded: that in his first week he found Figure had no formal safety procedures, no incident-reporting system and no risk-assessment process. That impact testing in July 2025 produced a calculation showing the robot generated striking force more than twice what is necessary to fracture an adult human skull, capable of severe permanent injury. And separately, that a robot malfunctioned and struck a refrigerator, leaving a gash about a quarter of an inch deep in the door, with an employee standing nearby who was nearly hit. He further alleges leadership approved safety roadmaps and then downgraded them, and that presenting those roadmaps to investors was potentially fraudulent. He was terminated on 2 September 2025, with the stated reason being a change in business direction.

The precision point: the skull figure is a force calculation from a test, and the refrigerator is a separate malfunction incident. No injury to any person is alleged in either. The widely shared version that merges them into "a Figure robot fractured someone's skull" is not what the complaint says.

Figure's response is that Gruendel was terminated for poor performance and that the claims are falsehoods it will discredit in court. As of 30 August 2026 we found no reporting of a ruling, settlement or dismissal - the last docket activity we could source was from January 2026. Treat it as pending and unresolved, and do not imply an outcome in either direction.

Why it belongs on a buyer-facing page: it is the clearest public illustration of why "when can this be in a home?" is a certification question rather than a software question.

Is there a Figure 03 SDK or developer program?

No. There is no public SDK, no developer API, no documentation, no simulator release and no third-party access of any kind. We looked, and the absence is complete across Figure's own materials and the 2026 industry SDK roundups.

Helix and Helix 02 are proprietary and run onboard. There are no published weights, no cloud endpoint and no inference API. The models are not available to anyone outside Figure.

One thing is genuinely open, and it runs in the opposite direction to what developers want. On 25 August 2026 Figure launched Index, a crowdsourced platform where members of the public record human video that Figure uses as training data and get paid for it. Figure reported 264,000 app downloads across 108 countries, more than 44,000 weekly active users, over 16 million videos, video arriving at roughly 30 minutes per second, and $15 million paid to creators to date, with more than $1 billion committed to data and compute over the following twelve months. That is a data-acquisition programme, not a developer platform. It lets you contribute to Figure's model; it does not let you build on Figure's robot.

For a robotics team that needs to write code against a humanoid in 2026, Figure is not an option and shows no sign of becoming one. That market belongs to Unitree, whose SDK, ROS 2 support and MuJoCo simulator are all public.

What would owning a Figure 03 actually involve?

Nobody owns one, so this section is about the shape of the cost rather than the numbers themselves.

Figure's deployment model is not a sale. Robots go into partner facilities with Figure engineers resident on site - that is how both BMW Spartanburg and the Catalyst Brands programme in Reno are structured. So the cost profile of a Figure deployment looks much more like a managed service contract than a capital purchase: integration, on-site engineering, iteration on the task, and ongoing model updates.

What that means practically for a prospective enterprise customer is that the hardware price is close to irrelevant. You are buying Figure's engineering time and Helix's task performance on your specific workflow, and the length of the integration determines the bill.

On the physical side, Figure has published enough to sketch the infrastructure. Charging is wireless and inductive through coils in the feet at 2 kW, which means a charging pad rather than a docking connector - simpler to install than it sounds, but it is a fixed installation. Five hours of runtime at peak performance means a single unit does not cover a two-shift day without a charging window. The 10 Gbps mmWave link is for data offload rather than control, so it implies local network infrastructure to receive it.

There is no published warranty, no service-level agreement, no spare-parts catalogue, no repair pricing and no published downtime figure. If you are modelling a deployment, those all become negotiated contract terms rather than list items, and you should assume Figure will not publish them.

Who is actually using Figure 03 today?

Two named commercial customers, plus Figure's own facilities — that's it. That is the complete public list.

BMW Manufacturing, Plant Spartanburg, South Carolina. Figure 03 on logistics sequencing in Assembly Hall 52, confirmed by BMW's release of 25 June 2026 and still reported active in late August 2026. Unit count undisclosed. This follows the Figure 02 body-shop deployment that ran roughly eleven months into November 2025.

Catalyst Brands, Reno, Nevada. Announced 26 May 2026, deploying at the company's distribution logistics centre to automate physically demanding supply-chain tasks. Catalyst is the parent of JCPenney, which confirmed the agreement through its own corporate newsroom the same day. No operational metrics have been published since - no throughput, no uptime, no unit count. Analysts have fairly described it as a signed pathway rather than a proven deployment, in contrast to BMW where runtime data exists.

There is a persistent rumour of a major US logistics customer, with UPS named in reporting from April 2025 as being in talks and a potential requirement running to 100,000 robots by 2029. Neither UPS nor Figure has ever confirmed it, and as of 30 August 2026 no such deployment has been announced. Treat it as unconfirmed.

Separately, and worth flagging because it gets confused: BMW also runs a humanoid pilot at its Leipzig plant in Germany using robots from a different vendor entirely. That is not Figure.

Figure 03 vs Tesla Optimus vs 1X NEO

Three robots at three genuinely different stages, and the ranking flips depending on what you measure.

On buyability, 1X leads. NEO takes consumer preorders at $20,000 or $499 a month with a $200 deposit, and reportedly took more than 10,000 orders in its first week. Figure has no price and no preorder. Tesla has no price and told analysts in August 2026 that external sales start no earlier than H2 2027.

On demonstrated autonomy, Figure leads clearly. The May 2026 package-sorting run - roughly 101,000 packages, 81 hours, with Adcock stating no teleoperation - is the strongest autonomy evidence any of the three has produced. By contrast, 1X openly builds remote human operators into its product as "Expert Mode", and Tesla's headline manipulation demos have all been human-driven, including the 2024 We, Robot event and the Tesla Diner unit.

On real deployment, Figure leads again. BMW Spartanburg and Catalyst Brands are named, dated, external customer sites. Tesla has none. 1X has an industrial agreement with EQT covering up to 10,000 units across portfolio companies through 2030, but no verified consumer delivery.

On manufacturing, Figure claims one robot per hour off BotQ as of April 2026 with 350+ delivered; Tesla was still installing its first line in July 2026. Both shipped in the low hundreds during 2025, roughly ten times behind Unitree.

On valuation, Figure was marked at $39 billion at its Series C close in September 2025 and has not been re-marked publicly since. That is a private negotiated round price, not a market price.

The blunt summary: Figure is furthest along on doing work, 1X is furthest along on selling to people, Tesla is furthest along on nothing measurable yet but has the deepest resources. Nobody has put a humanoid into a private customer's home.

Who should be watching Figure 03, and who shouldn't bother?

Watch it closely if you run logistics, warehousing, distribution or manufacturing operations at scale. Figure is the humanoid company with actual production deployments at a named automaker and a named retailer, and its stated model - embedded engineers, task-specific integration, robots as a service rather than a purchase - is the one most likely to reach your operation first. If you have repetitive, physically demanding sequencing or sorting work, Figure is worth a conversation, and the conversation starts with Figure's enterprise team because there is no other channel.

Watch it as an investor or analyst too, with the caveat that you cannot act on it. Figure is private, was marked at $39 billion in September 2025, and explicitly voided unauthorised secondary sales in July 2026.

Do not wait for it if you are a consumer. There is no home product, no price, no preorder, no certification and no 2026 update on the home alpha programme. Adcock's $400-$600 monthly lease sketch from May 2026 has no start date attached. If you want a home humanoid this decade, 1X NEO is the only one taking orders, and it has not delivered one either.

Do not consider it if you are a developer or a research lab. There is no SDK, no API and no model access, and nothing suggests that changes. Buy a Unitree G1 EDU or an R1 EDU instead - they cost real money and you can actually program them.

And if you are writing a plan that depends on Figure availability, anchor on the two things that are documented: BMW Spartanburg is real and running, and Figure has never sold a robot to anyone.