526 million tasks, two dollars each: Pi’s human workforce

Pi Network paid more than a million people to verify eighteen million identities across half a billion discrete tasks, then published the exact formula for what it paid them. Every outlet reproduced the formula. Not one converted it to dollars. Here is the number, and what it says about crypto’s largest experiment in distributed human labour.

There is a specific kind of number that gets repeated across dozens of articles without anyone stopping to convert it, and Pi Network produced a textbook example this year. The project announced that more than a million verified users had completed over 526 million identity-validation tasks, confirming eighteen million people across two hundred countries, and that all of those validators had now been paid. It is a remarkable operational achievement, arguably the largest distributed human-labour experiment ever run on a blockchain, and Pi has since positioned that workforce as infrastructure available to artificial intelligence companies that need verified humans in the loop. The coverage was extensive and uniformly impressed.

Pi published the exact reward formula on its own blog. Eight or more outlets reproduced it faithfully, quoting 0.0504179 Pi per validation and noting that this represents 21 times the base mining rate. One publication’s page carries the sentence “arriving at a price per validation of approximately” followed by nothing at all, the figure simply missing. Nobody multiplied by the token price. This piece does, and the resulting number reframes both the achievement and the business built on top of it.

The formula, and the number it produces

Start with Pi’s own arithmetic, because the project publishes it in full and there is no dispute about the inputs.

Every Pioneer who completed identity verification and migrated to the Mainnet contributed 1 Pi into a shared reward pool. By the snapshot date of March 5, 2026, that produced a pool of 16,568,774 Pi, corresponding to the same number of migrated users. The Pi Foundation added a further 10 million Pi, acknowledging that much of the earliest validation work had served to train the validator workforce instead of processing final applications. Total pool: 26,568,774 Pi.

That pool was divided by 526,970,631 successful validations, producing a price per validation of 0.0504179 Pi. Validators needed at least 50 qualifying validations reaching majority agreement to receive a payout, and rewards were transferred directly to Mainnet wallets.

Now the conversion nobody performed.

At Pi’s current price of approximately $0.077, the entire distribution is worth roughly $2.05 million. A single validation pays about $0.0039, four tenths of one cent. The minimum qualifying threshold of 50 validations pays approximately $0.19.

Divide the task count by the validator count and the average participant completed around 481 validations, earning roughly 24.3 Pi. At current prices that is about $1.87. At the price when the rewards were distributed in early April, when Pi traded near $0.176, it was closer to $4.27.

So the headline reads: more than a million people completed half a billion tasks and received, on average, somewhere between two and four dollars each.

What the multiple actually means

The most-repeated framing in the coverage is that the rate represents 21 or 22 times the base mining rate, and Pi states this directly. It is accurate, and it is worth thinking about what it implies instead of accepting it as the reassurance it is presented as.

If 0.0504179 Pi is 21 times the base mining rate, the base mining rate is approximately 0.0024 Pi. At current prices that is roughly two hundredths of one cent.

A multiple is only informative relative to its base. Twenty-one times a very small number is a slightly less small number, and describing the validator rate as 21x invites the reader to conclude that validation is well compensated, when what it actually shows is that mining rewards are worth almost nothing and validation is worth somewhat more than almost nothing.

This is the same pattern this publication has documented elsewhere in crypto metrics. A network celebrating 1.4 million agent transactions turned out, on the arithmetic, to have generated roughly $280 in total fees, because transactions on that chain cost two hundredths of a cent. In both cases the underlying activity is real, the technology works, and the number that gets promoted measures volume while the number that would measure value goes unpublished. Our guide to why transaction counts mislead covers the general form of the error; this is the labour-market version of it.

Why the task count is so large

One structural point deserves explanation, because it is the reason 18 million identities required 527 million validations and it is to Pi’s credit.

Pi’s verification system is built for privacy preservation. Rather than handing a complete application to one reviewer, the system splits each application into discrete tasks: a liveness video check, a document review, a photo match, a>The AI business built on this number

The reason the arithmetic matters is that Pi is now marketing this workforce commercially, and the economics of that offer depend entirely on what the workforce costs.

Pi has formally articulated a strategy positioning its verified user base as infrastructure for artificial intelligence companies, targeting data labelling, reinforcement learning from human feedback, and model evaluation. The pitch is specific and, on its own terms, strong: more than 18 million identity-verified humans across 200-plus countries, each already holding a functioning wallet, with a proven production-scale precedent of half a billion completed tasks. Both co-founders took that pitch to a major industry conference this year.

The strategic logic is real. Verified-human labour is genuinely scarce and getting scarcer as generated content and automated accounts contaminate every open data source, and the AI industry does need what Pi has built. Existing distributed-work platforms struggle with exactly the problems Pi has solved: onboarding friction, payment rails across jurisdictions, and confidence that the worker is a person.

The uncomfortable half is the price. A workforce that accepted four tenths of a cent per task, denominated in a token trading near its all-time low, is either an extraordinary cost advantage or an unsustainable one, and which it is depends on why people participated. If validators were working for the token’s future value and not its present value, the labour supply is contingent on price expectations that the chart has been steadily contradicting. If they were working because the task volume was low and the marginal effort trivial, the supply may not scale to commercial>

The gap in the user base

One further figure deserves the same treatment, because it sits alongside the task count in every Pi announcement and receives even less scrutiny.

Pi describes an engaged user base in the tens of millions, with figures above 60 million appearing routinely in its communications and in coverage. The KYC programme this piece examines verified approximately 18 million identities. Of those, 16,568,774 had migrated to the Mainnet by the March snapshot, which is the figure the reward pool was built from, since each migrated Pioneer contributed exactly 1 Pi.

Set those numbers against each other and the funnel is stark. Something on the order of 60 million people engaged with the application. Roughly 18 million completed identity verification. Roughly 16.6 million completed migration to the Mainnet. The gap between the top and bottom of that funnel is larger than the entire verified user base of most cryptocurrencies.

The attrition is not necessarily damning, and the reasons are mundane, not sinister. Verification requires documents many users do not have or will not submit. Migration requires deliberate action from people who joined a free mobile application years ago and may have forgotten it. Some accounts were duplicates or automated, which the verification process exists to catch, and catching them is a success, not a loss.

But it matters for the commercial pitch, because the number Pi markets to potential AI customers is the verified figure and the number that appears in most public discussion is the engagement figure. Those are different populations by a factor of more than three, and the addressable workforce is the smaller one. Of that smaller one, roughly a million people, about six percent of verified users, actually performed validation work in the first round.

A workforce of a million active participants is substantial and unusual for a blockchain project. It is also considerably smaller than the headline suggests, and any assessment of what Pi can deliver commercially should start from the million who worked, not the sixty million who once downloaded something.

What the number does not prove

An honest treatment names what the arithmetic cannot settle, and there are three things.

It does not prove exploitation. Validators opted in voluntarily, the work was intermittent and required no commitment, most participants were already mining Pi on their phones for years at rates the validation payout exceeded twenty-fold, and nobody was induced to leave other employment. Comparing four tenths of a cent per task to a minimum wage assumes an employment relationship that did not exist.

It does not prove the workforce is worthless. Half a billion completed tasks with majority-agreement thresholds is a real operational output, and the fact that it was cheap says as much about Pi’s ability to mobilise its community as about the rate. Any conventional provider attempting the same verification volume with the same privacy compartmentalisation would have paid enormously more.

And it does not settle the token question. The value of these rewards is a function of Pi’s price, which sits near its all-time low after a long decline. The same 24 Pi that is worth under two dollars today was worth more than four dollars in April and would be worth considerably more if the token recovered. Validators paid in an asset instead of cash hold a claim whose value is undetermined, which is the oldest arrangement in crypto and cuts both ways.

What the number settles is the framing. An achievement described in hundreds of millions of tasks is, in dollar terms, a two-million-dollar programme, and any assessment of Pi’s commercial AI ambitions should start from that figure rather than from the task count.

The comparison the pitch invites

If Pi is selling a verified-human workforce to artificial intelligence companies, the honest exercise is to price it against what those companies currently pay, because that comparison is the entire commercial case and nobody has run it.

The established distributed-labour platforms operate on per-task pricing that varies enormously by complexity, from fractions of a cent for the simplest classification work to several dollars for tasks requiring judgment, domain knowledge, or sustained attention. Reinforcement learning from human feedback, the category Pi names explicitly, sits toward the expensive end, because it requires annotators who can evaluate model outputs coherently and consistently, and the labs buying it have historically paid accordingly.

Pi’s demonstrated rate is four tenths of a cent per task. That is competitive at the very bottom of the market and nowhere near the categories Pi is targeting in its pitch.

Two readings follow, and they point in opposite directions. The optimistic one is that the rate reflects what Pi chose to pay for internal work using a token it issues, not what it would charge a commercial client, and that a paying customer’s budget would flow to validators at market rates with Pi taking a spread. On that reading the 526 million tasks prove capability and mobilisation, not price, and the low figure is irrelevant to the commercial offer.

The sceptical reading is that the participation itself was a function of the rate being incidental. People completed validations in spare moments, on phones, for a token they were already accumulating, with no expectation that the payout would matter. Asking the same population to perform sustained, quality-controlled annotation work for real money is asking for a different behaviour from a different worker, and the fact that a million people tapped through simple checks says relatively little about whether fifty thousand of them will label training data to a standard a laboratory accepts.

Both readings are consistent with the evidence, and the distinguishing test is straightforward: a named commercial client, a task type, and a rate. Until one of those exists, the 526 million figure proves that Pi can mobilise its community for near-free work, which is a genuine and unusual asset, and does not yet prove that the community will work for customers.

What Pi built that others could not

Setting the arithmetic aside for a moment, an honest assessment has to credit what this programme accomplished, because the achievement is real and the criticism above does not touch it.

Verifying eighteen million identities across more than two hundred countries is a task that established identity providers charge substantial sums for and frequently perform badly. The industry standard is either full automation, which fails on document variety and produces false rejections at scale in exactly the countries with the least standardised paperwork, or outsourced human review, which concentrates sensitive personal data in a small number of processing centres and has produced repeated breaches. Pi did neither. It built a system that splits each application across many reviewers so that no individual sees a complete file, requires independent agreement at each step, and pays participants in the network’s own asset through infrastructure it also built.

That combination has not been achieved elsewhere at this scale, and the privacy property in particular is a genuine engineering accomplishment, not a marketing claim. Compartmentalised review is harder, slower, and more expensive in labour terms than the alternatives, which is precisely why nobody does it, and Pi absorbed that cost by having a community willing to work for a token.

The payment rail matters too and receives even less attention than the verification design. Distributing rewards to more than a million people across two hundred countries, in amounts averaging a couple of dollars, is operationally impossible through conventional financial infrastructure. Payment processors will not handle it, the fees would exceed the payments, and the compliance burden of onboarding a million micro-payees in that many jurisdictions would swamp any project attempting it. A blockchain with pre-verified wallet holders solves that specific problem completely, and it is the clearest instance in this whole story of crypto doing something the existing system genuinely cannot.

So the fair summary holds two things at once. The economics are far smaller than the headline numbers imply, and the infrastructure that produced them does something no conventional provider could. Whether the second fact can be sold to customers at a price that makes the first fact irrelevant is the entire question hanging over Pi’s commercial strategy, and it will be answered by a contract rather than by a milestone announcement.

What to watch

The second distribution round. Pi says it is refining the validator performance algorithm before the next round and expects higher per-validation rates. Whether that materialises, and at what dollar value, is the single most informative upcoming data point.

Whether any AI customer signs. The commercial pitch has been made publicly. A named client, a contract value, or a disclosed pilot would convert the strategy from a positioning statement into a business. Its absence over the coming quarters would be equally informative.

Task volume after automation. Pi expects automation to reduce human validations per application. That improves per-task pay and shrinks the total workforce opportunity simultaneously, and the two effects pull in opposite directions for anyone valuing the labour network.

The token price against participation. If validator participation holds while the price falls, the labour supply is not primarily price-motivated, which strengthens the commercial case considerably. If it falls with the price, the workforce is a function of speculation rather than of wage.

Open Mainnet. The transition remains pending with no announced date, and the terms of it determine whether validators can convert rewards freely, which is what makes any of these numbers real for the people who earned them.

A closing note on why this conversion was worth doing at all, since the arithmetic is trivial and the inputs were public the whole time.

Crypto produces an unusual density of figures that are technically accurate and practically uninformative, and they share a shape. A count, a multiple, or a cumulative total, quoted without the unit that would let a reader size it. Half a billion tasks. Twenty-one times the base rate. Sixteen and a half million Pi plus ten million more. Every one of those statements is true, verifiable, and published by the project itself in good faith. Together they produce an impression of scale that a single multiplication dissolves.

The reason the multiplication does not get done is not conspiracy. It is that the party with the strongest incentive to publish the number is the party with the least reason to, and the outlets covering the announcement work from the press release under time pressure, reproducing the figures they are given. One of them, working from Pi’s own blog post, left the per-validation figure out of the sentence entirely and published anyway. That is what a supply chain of unconverted numbers looks like in practice.

The correction is available to any reader with a calculator, and the habit generalises well beyond this project. When a crypto announcement leads with a count, find the price and multiply. When it leads with a multiple, find the base. When it leads with a cumulative figure, find the period. The result is frequently smaller than the headline suggests, occasionally larger, and always more useful than the number you were handed.

Frequently Asked Questions