Figure AI’s Index App: What’s Verified, What’s Not, and How It Works

Figure AI has taken its robot-training data platform out of stealth. On August 25, 2026, the humanoid robotics company announced that Index — a mobile app that pays people to record videos of everyday physical tasks — is now generally available on the App Store and Google Play, after four months of quiet testing.

The pitch is straightforward: general-purpose robots can’t learn to move through kitchens, warehouses, and offices from text scraped off the internet the way language models learn to write. They need real footage of real people doing real things. Figure’s answer is to turn ordinary phones into a global data-collection network, and pay the people using them.

What makes this moment worth examining isn’t just that the app exists — it’s separating what Figure has actually shown from what it’s only claimed. Here’s what the evidence supports.

What Is Index?

Index spent four months in a stealth testing phase before Figure launched it publicly under the Index name. On its own announcement page, Figure describes the project as “the largest useful robot training dataset in the world” — a characterization worth flagging up front as the company’s own framing rather than an independently confirmed superlative. No outside benchmark or audit currently supports or disputes that ranking.

The data collected through Index feeds Helix, Figure’s vision-language-action AI system, which is the software that controls the company’s Figure 03 humanoid robots.

Why This Matters for Robotics

Figure’s stated reasoning is that text-based AI had an enormous advantage: the internet is essentially a pre-built training corpus. Robotics has no equivalent. There’s no large-scale public archive of how a human hand adjusts its grip on an unexpectedly heavy coffee cup, or navigates a cluttered laundry room. That kind of physical, situational data has to be captured from the real world, one recorded task at a time.

Figure has described building a Figure-exclusive collection pipeline as necessary to get real-world physical-task footage at the throughput, diversity, and quality it says the Helix model requires. That’s Figure’s own account of its strategic rationale, not an independently verified assessment.

How Index Works

Index operates less like a typical data-labeling app and more like a two-sided labor marketplace. Contributors, whom Figure calls “Creators,” can either:

• Record themselves performing everyday tasks — cooking, cleaning, doing laundry — using the app, and get paid for it, or
• Be booked through the app by a person or business who wants help with tasks like stocking shelves, prepping food, or general chores, with the recording happening as a byproduct of the work itself.

Figure says it has captured tasks in both home and commercial settings, including logistics centers, restaurants, factories, and offices — with company materials citing examples as specific as cleaning kitty litter, changing oil, and busing restaurant tables.

On the hardware side, Figure’s public materials describe a smartphone-based recording flow, with an option for Creators to “receive a recording device” upon signing up. Figure has not published a specification sheet detailing what that device is, what phone models are supported, or any minimum camera requirements — so specifics beyond “smartphone plus an optional company-provided device” are not publicly available.

The Data Pipeline

Raw footage from millions of contributors isn’t automatically useful. Figure describes a five-stage curation pipeline that incoming video passes through before it becomes training data:

1. Filtering — automated checks for technical, visual, and semantic quality
2. Fraud review — human analysts audit samples to catch attempts to game the system
3. Deduplication — video segments are embedded and near-duplicates above a similarity threshold are discarded
4. Rebalancing — submissions are sorted using task quotas and embedding-based clustering to preserve variety
5. Annotation — every accepted clip receives hierarchical text captions describing the action

This is a real, described process — but it’s a company account of its own quality-control system. No third party has published error rates, accuracy figures, or an independent audit of how well the pipeline actually screens out low-quality or fraudulent submissions.

It’s also worth distinguishing two things Figure’s numbers technically say. The company states that Creators have uploaded more than 16 million videos — that’s different from confirming that all 16 million have fully cleared the five-stage pipeline and become usable training data. Figure doesn’t separately disclose how much of that footage is fully processed versus still in the pipeline.

The Numbers Figure Is Reporting

At launch, Figure shared the following figures, all sourced to its own announcement:

These figures come directly from Figure’s launch announcement and its Index landing page. As of this writing, weeks after the public launch, the same figures still appear on Figure’s site without any visible change — which may simply mean growth has been modest since launch, or that the on-page counters aren’t updated in real time. Either way, it’s a detail worth keeping in mind rather than assuming the numbers reflect the current, live state of the platform.

The $1 Billion Commitment — Context and Caveats

Figure also stated it is “committed to spend over $1B the next 12 months on data and compute.” This is a forward-looking pledge, not a confirmed amount already spent on Index specifically.

There is independent context that makes such a commitment financially plausible: in September 2025, Figure announced it had raised more than $1 billion in a Series C funding round at a $39 billion post-money valuation, led by Parkway Venture Capital with participation from Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, and other investors. That raise was independently reported by outlets including Bloomberg, separate from Figure’s own announcement.

That funding round establishes that Figure has recently had access to capital at the scale needed to back a billion-dollar commitment. It does not, however, independently confirm that the specific $1 billion pledged for data and compute has been or will be spent as described.

What’s Not Publicly Disclosed

Several questions a prospective contributor or industry observer might reasonably ask remain unanswered in Figure’s public materials:

• Payout terms. Figure has disclosed an aggregate $15 million paid to Creators, but not the per-video or per-minute rate, payment frequency, or contract terms individual Creators agree to.
• Data ownership and licensing. Figure’s general Privacy Policy, which applies site-wide, notes that the company collects “Sensory Data such as photos, videos, or recordings of you and/or of your environment.” But there is no separate, publicly posted Creator agreement addressing who owns the footage, how it can be reused or resold, or what licensing rights Creators retain.
• Hardware specifications. Beyond the smartphone-based flow and an optional recording device, no spec sheet is available.
• Geographic completeness. Figure has not disclosed whether any countries or demographic groups are excluded from its “100+ countries” figure.
• The underlying research method. No published technical paper describes how Figure converts ordinary 2D smartphone video into the 3D motion data a robot needs to learn from.

None of this means the answers don’t exist — only that they aren’t publicly available as of this writing.

Limitations and Unknowns

The most important caveat across this entire story is source concentration: nearly every headline number — downloads, uploads, payouts, the billion-dollar pledge — traces back to a single company announcement. That’s not unusual for a product launch, but it means readers should treat these as Figure’s own reported figures rather than externally audited facts.

Figure has also said internally that “generalization results” are already validating its data-first thesis for robotics. As of now, no benchmark, paper, or independent evaluation has been published to support that claim.

Conclusion

What can be said with confidence: Index is a real, live product. It’s downloadable today on both major app stores, it pays people for contributing footage, and it’s tied to a company that recently raised substantial capital. The scale figures Figure has shared — hundreds of thousands of downloads, millions of videos, $15 million in creator payouts — describe a genuinely large operation, at least as the company presents it.

What remains open: independent verification of those numbers, the actual terms Creators sign up for, what rights they retain over their own footage, and whether this approach measurably improves how well Figure’s robots generalize to new environments. Those are the questions worth watching as Index moves from a stealth experiment to a public product with real financial and personal stakes for the people contributing to it.

FAQ

What is Figure AI’s Index app? Index is a mobile app from humanoid robotics company Figure AI that pays people to record video of everyday physical tasks — like cooking, cleaning, or warehouse work — which is then used to train Figure’s Helix AI system.

How does Figure AI pay people to use Index? Figure has disclosed that it has paid Creators $15 million in aggregate since the app’s stealth phase began. The company has not publicly disclosed specific per-video or per-minute payment rates.

Is the Index app available in my country? Figure states the app has users across 108 countries (also described as “100+” in its announcement). It has not published a specific list of supported or excluded countries.

What does Figure AI do with the videos people upload? Uploaded footage passes through a five-stage process — filtering, fraud review, deduplication, rebalancing, and annotation — before becoming training data for Helix, Figure’s AI system for its humanoid robots.

How much has Figure AI actually paid Index Creators? Figure states $15 million paid to Creators as of its August 2026 launch announcement. This figure has not been independently audited.

Is Figure AI’s $1 billion data commitment already spent? No. Figure describes this as a commitment to spend over $1 billion on data and compute over the next 12 months — a forward-looking pledge, not money already spent.

What hardware do I need to become an Index Creator? Figure’s public materials describe a smartphone-based recording process, with an option to receive a recording device upon signing up. Detailed hardware specifications have not been published.