2026 H1 Report: Flash Boys or Superforecasters?
A deep dive into Polymarket’s venue growth, user portraits and market structures
Who actually trades prediction markets? Six months of complete on-chain activity and profiles — every fill, every wallet: how users arrive, what keeps them, who supplies the liquidity, and where the money flows.
Based on 787 million deduplicated fills across 1,733,011 wallets on Polymarket’s international exchange, Jan–Jun 2026, plus venue-level data for Polymarket US.
- A second Polymarket is scaling fast. The CFTC-regulated US venue grew from 5% of the international exchange’s monthly volume in January to 47% by June — and 67% by July.
- High-turnover categories are the fee engine. Sports and Crypto turn capital fastest — and print 98% of the international venue’s $184M in H1 fees, with Crypto ($96.3M) out-earning Sports ($83.5M) on half the volume.
- 90% of users are retail; while bots & pros do 80% of the volume. Bot-alike wallets — 5.9% of the user base — take two-thirds of all volume alone; with the informed/pro cohort they account for nearly three-quarters.
- User acquisition is diversified, not hit-driven. The single biggest onboarding market brought in under 7K of H1’s 1.13M new wallets, and a month’s top-15 capture just 8–19% of its intake.
- Loyalty accrues to the platform, not the category. ~51–57% of actives return each month, while the entry category barely moves retention (44–49% at one month nearly everywhere) — users churn out of categories but stay on Polymarket.
- Bots win as a class — but the median loses money. The bot cohort nets +$108M, yet only 27% of bot wallets profit and the median bot loses; just 146 of 1.75M wallets were profitable all six months.
The Rise of Polymarket US
Polymarket spent H1 2026 becoming two exchanges — and the discourse only watched one of them. The original venue — global for non-US customers, an off-chain CLOB with on-chain settlement on Polygon — Polymarket International cleared $17.9B of one-sided volume in six months and held stable through the summer, even as commentators read the headline chart as a loss of ground to Kalshi.
The quiet part of the story is its younger sibling: the CFTC-regulated Polymarket US, launched in late 2025, went from 5% of the international exchange’s monthly volume in January to 47% by June; and 67% by July ($2.0B vs $3.0B). A second Polymarket is growing up fast.
The two share a brand but not order books, not a user base, and — as we will see — a very different composition of market catalog and revenue engine. What they do share, with each other and with Kalshi, is market design at a high level: all three venues run central limit order books with off-chain matching, charging takers on the same price-sensitive curve — fee = Θ·C·p(1−p). The international venue settles every fill on Polygon in public; Polymarket US clears the identical mechanics on a private CFTC-regulated ledger. Same engine, different rails.
A note on conventions before any cross-venue number: every venue’s headline volume follows its own definition. Throughout this report, volume refers to single-leg, taker-side dollars — as a rule of thumb, the notional and double-counted conventions run 2–3× higher than this measure. Kalshi reports contracts at $1 face value regardless of traded price (the notional-volume convention); Polymarket’s public feed logs every fill twice; similarly, the Polymarket US dataset’s default column is both-sides contract value. Put July on one measuring stick — one-sided dollars actually traded, counted once — and the conventional $41B / $7B / $5B become $13.3B / $3.0B / $2.0B.
Pantera Research Lab@sui414 · @0xallyzach
The US venue’s volume compounds from ~$1M a day at launch to $50–100M a day by July; transactions cross a million a day; open interest peaks above $150M. Fees switch on in January at a flat basis point, then move in April to the price-sensitive formula both venues now share.
Pantera Research Lab@sui414 · @0xallyzach
The fee engine is real, and bigger than the community datasets suggest: $30M of gross taker fees in June alone. Working through the official rebate schedule — a 25% maker rebate at the point of trade, a promotional all-taker rebate that expired April 30, then volume-tiered taker rebates — puts modeled retained revenue near $17M for June, roughly double the flat-25% assumption most trackers carry. One important clarification for this estimation is that the share of fees from large takers is modeled referencing the international venue’s mix, where the top 0.3% of takers do ~63% of the volume, since we can’t identify who qualifies for tiers due to the lack of user-level data for the US venue.
Fee Economics on Polymarket International
What is the world betting on? On Polymarket International, half of everything traded is sports. Crypto follows at roughly 30% — overwhelmingly five-minute up-or-down micro-markets, though its single largest market is the controversial “MicroStrategy sells any Bitcoin by May 31” ($95M). World/Geopolitics ranks third at ~15%. What varies more than volume is concentration: Sports’ biggest market takes up only 0.27% within its category, while a single “will the US confirm aliens exist” market is a quarter of Science & Tech.
Pantera Research Lab@sui414 · @0xallyzach
The international venue generated roughly $184M in H1 gross fees — not net revenue, since maker rebates and incentives come out of that. Category economics differ sharply: with only half the volume, Crypto ($96.3M) printed more fees than Sports ($83.5M), and together the two account for 98% of it. World/Geopolitics, third by volume, runs deliberately fee-free.
Pantera Research Lab@sui414 · @0xallyzach
High Turnover Category as the Fee Engine
In July, Kalshi held ~2.3× Polymarket’s open interest but did ~4.4× its volume — implied daily turnover of ~0.44× against ~0.23×, with Polymarket US at ~0.64× actually turning its capital fastest of the three. So how does a venue holding barely 2× the parked capital print 4× the flow — and why is the fastest-turning book of all the smallest one?
Pantera Research Lab@sui414 · @0xallyzach
Pantera Research Lab@sui414 · @0xallyzach
The answer lies in the structure of the market catalog. Volume and open interest measure different aspects of the business, and the daily turnover ratio — dollars traded over collateral locked — reveals how differently the market categories behave.
Category by category: Sports and Crypto are the fastest-turning categories, while World / Geopolitics is the slowest, at just 0.01–0.03× daily turnover. This means around $200–300M of capital sits in open interest — for both Polymarket International and Kalshi — but prints only $2–3M of volume, and both venues set the category fee-free. Instead, with 88% of open interest sitting in the Sports category for Polymarket US, it explains why, as the smallest venue, it has the highest turnover rate.
Kalshi’s crypto is the true outlier, turning its open interest nearly 6× a day — $96M of daily flow against $17M of capital, 98% of it in markets that live an hour or less (e.g., betting on whether the BTC price is up or down in 15 minutes or 1 hour).
Polymarket’s crypto prints just 0.31× because its open interest sits in a long-dated shelf its micro-markets don’t touch: the average 5-minute up-or-down market only ever locks about $2,800 of collateral — the markets that do 90% of the category’s volume hold essentially none of its capital, turning over dozens of times a day. The open interest lives elsewhere — in a shelf of dated threshold markets (a ladder of “will Bitcoin dip to $35–55K by December 31” insurance strikes, year-end lottery tickets like “Bitcoin $100K,” and, the single largest holder at $1.4M, “Will Satoshi move any Bitcoin in 2026”). Buy once, wait for December.
One other observation is that the same category converges to the same speed across platforms — sports clears 0.57× a day on Kalshi, 0.67× on Polymarket International, and 0.64× on the US venue — roughly 0.6× a day on all three venues, which says turnover is an attribute of the market category rather than the venue.
Discussion below on user base only covers Polymarket’s international exchange (Polygon CTF/NegRisk contracts), Jan 1 – Jun 30, 2026. Polymarket US (QCX LLC, a CFTC-designated contract market) clears on an internal ledger through FCM intermediaries. It leaves no on-chain footprint, its public trade stream is anonymized to side-and-intent by design, and no leaderboard or profile surfaces exist. Same applies to Kalshi.
Maker vs Taker Split
1,733,011 wallets traded on the international venue in H1. But the two sides of the order book are not symmetric populations — nearly every wallet has taken liquidity at least once, while fewer than half have ever rested a limit order. The maker side is the professional side, and it is small. The top thousand makers provide the majority of all resting liquidity. On the demand side, six percent of wallets — the hyperactive tier — do four-fifths of the taking, while a third of all wallets place five or fewer trades in six months.
Pantera Research Lab@sui414 · @0xallyzach
User Growth Accounting
User growth tells a less flattering story than volume: monthly active wallets were nearly flat across the half — 580K in January to 594K in June — despite 1.1M wallets trading for the first time. Each month, 43–49% of the prior month’s active wallets stopped trading, with churn peaking at 358K in April, forcing the platform to continually refill the funnel with new and resurrected wallets. Acquisition was strong — peaking in March — but active-wallet retention is the bottleneck.
Three caveats before reading too much into that. A wallet is not always a user. Some “churn” is users rotating into fresh wallets. And churn here means a wallet stopped trading — not that an account was closed or unfunded, the way a fintech like Robinhood would define it in a filing.
In the chart below, each month’s actives split into retained, resurrected and new wallets; red bars below the axis are prior-month actives who left, as churned users. March’s peak drew record new and resurrected users. April shed 358K — the half-year’s biggest churn. May’s floor was propped up more by returners (126K) than by new arrivals (116K).
Pantera Research Lab@sui414 · @0xallyzach
Where New Users Come From
Attribute each new wallet to the first market it ever touched and the acquisition machine becomes visible. Three categories do almost all the onboarding, each with its own season: crypto peaking at 99K new wallets in March, geopolitics at 68K in January, sports at 91K in June — the World Cup’s opening month. And no single market did it all: major events bring waves (January’s Iran headlines, June’s Taiwan market and the World Cup), yet even the top onboarding market brought in fewer than 7K of H1’s 1.13M new wallets.
Pantera Research Lab@sui414 · @0xallyzach
The onboarding moments are dispersed. Even the #1 onboarding market brought in under 7K of 1.13M new wallets. Listings win in aggregate, not through a single hit.
Pantera Research Lab@sui414 · @0xallyzach
Similarly, the top-15 onboarding markets of a month capture just 8–19% of its new wallets; the #1 market alone never exceeds ~3%. The two most concentrated months are event-driven: January (Iran headlines) and June (the Taiwan market and the World Cup).
Pantera Research Lab@sui414 · @0xallyzach
Retention by Market Category
About half of any month’s active wallets trade again the following month — the platform retains ~51–57% month to month; but loyalty to a category is far weaker than loyalty to the platform. Users churn out of the market type that brought them in and reappear somewhere else — this potentially says the cross-sell engine is at work.
Pantera Research Lab@sui414 · @0xallyzach
Comparing across categories, the type of entry market does not seem to diverge in retention rate. Retention clusters tightly across nearly all categories — 44–49% of new wallets remain active after one month, 25–32% after three — regardless of the door they came through. The one clear outlier is Crypto, at a 37% retention rate after one month — entrants retaining roughly 30% lower than every other category’s. One explanation is that the automated operations of the crypto trading category have a preference to rotate out new wallets from time to time, which lowers the retention number on paper.
Pantera Research Lab@sui414 · @0xallyzach
The User Base Matrix
The matrix below provides a view for tracking user footprints across market categories and tells us about the platform’s cross-sell engine: the diagonal numbers are the absolute count of total unique wallets that traded in each category, and the % value — in every pair of categories — represents the share of the smaller category’s users who also trade the other.
Sports × Crypto overlap is only 49% — despite being the two largest user groups, they are two large but meaningfully distinct audiences, with Sports users expanding into other categories more readily than Crypto users, who tend to stay inside Crypto.
World/Geopolitics is the strongest connective layer, with consistently higher overlap against everything (except Weather) — evidence that these markets are a core product for the international venue, attracting the audience with the broadest interest across the platform.
Pantera Research Lab@sui414 · @0xallyzach
The Quadrants: Loyalty vs Expansion
Put onboarding, retention and cross-sell on one plane and every acquisition channel gets coordinates: loyalty — whether a category’s new users are still on the platform a month later — against expansion — how many other categories its users end up trading. Six of seven categories cluster at 43–49% loyalty, arranged along a spectrum from Sports’ loyal homebodies to Science & Tech’s engaged explorers, whose users average 3.6 other categories.
Crypto sits alone in the bottom-left: 37% loyalty and the least expansion. The five-minute casino acquires tourists — users who neither stay nor branch out. For a venue deciding what to list next, this is the acquisition-quality menu: sports buys retention, niche categories buy cross-sell, crypto buys volume that leaves.
Pantera Research Lab@sui414 · @0xallyzach
The Distribution of User Portraits
We depict the user base via a few different lenses: category breadth, trade size and trade frequency.
How specialized are they? (first tab) This is determined by how many of the seven market categories a wallet has traded: a pure specialist stays in one category, a semi-specialist touches two or three, a diversified wallet four to six, and a full generalist trades all seven. The result shows that category breadth correlates with size: as breadth expands, the average volume per wallet goes up — wallets that touch all seven categories are one percent of the population and trade seven times the volume per wallet of a pure specialist ($59K vs $8.3K average H1 taker volume per wallet).
How frequently do they place a trade? (second tab) This highlights the engagement level, ranging from a no-longer-active “tourist” to a hyperactive user. The funnel shows the 80/20 rule: six percent of “hyperactive” wallets do 79% of the volume, while one-fill tourists are eight percent of the base.
Pantera Research Lab@sui414 · @0xallyzach
For most of the categories, wallets with one or two fills dominate the user base — except in news markets. A single fill is the most common outcome in six of seven categories, highest in Sports (46%). World/Geopolitics inverts the pattern: only 20% single-fill — people come back to the same news market again and again as headlines move.
Pantera Research Lab@sui414 · @0xallyzach
The Three Tribes: Bots, Pros and Retail
Classify every wallet by how it behaves — frequency, sizing uniformity, breadth, ticket size — and the user base resolves into three tribes. Note that the heuristics are directionally indicative, but non-deterministic of the real characteristics behind each wallet’s identity. Thresholds are tested, but they can definitely affect the distribution results.
more than 100 fills per active day, or more than 500 fills with robotically uniform sizing (CV < 0.05), or more than 1,000 distinct markets
avg fill ≥ $500 with ≥ 5 fills across ≤ 50 markets, or avg fill ≥ $200 with ≥ 10 fills across ≤ 100 markets, or > $50K volume concentrated in ≤ 30 markets
everyone else — the default cohort
Bot-alike wallets — 5.9% of the user base — do two-thirds of the volume: probable bots — 102K wallets — took $11.9B of the $17.9B, and account for 89% of all fill events; while 1.54M retail wallets took $4.1B. Retail, under this heuristic, takes up 90% of the wallet user base. Informed/pro wallets are just 5% of the population and 6.6% of taker volume — consistent with a cohort that mostly rests orders rather than takes.
Pantera Research Lab@sui414 · @0xallyzach
By category, bots own the two biggest categories outright, while the contested middle — geopolitics, finance, culture — is where humans still set the price. Bot-alike wallets take 80% of Crypto and 72% of Sports volume — and 57% of Weather. World/Geopolitics, Finance, Culture and Sci-Tech split roughly three ways; meanwhile, informed/pro share peaks in World/Geopolitics (29%), Sci-Tech (28%) and Finance (25%).
Pantera Research Lab@sui414 · @0xallyzach
However, the volume dominance above comes from a thin bot layer present in every market type. By headcount, every category is overwhelmingly retail, with a noticeable drop in Crypto and Weather — a 5% share switched and filled up by bot actions.
Pantera Research Lab@sui414 · @0xallyzach
Are Polymarket Users Fee-sensitive?
Fees arrived at different times in different categories — this is a natural experiment where we can test the before-after effect, with fee-free World/Geopolitics as the control. It makes us curious: are prediction market users fee-sensitive? How does that differ by cohorts of users who trade bigger and more? And how does it differ by market categories?
The result below shows that the composition of volume shifts visibly around each category’s fee date, particularly for three markets: Crypto, Culture and also Weather — on all of which Informed/Pro volume specifically churned quickly after fees came on. Retail and bot volume stayed at the same share, showing as non-sensitive.
Pantera Research Lab@sui414 · @0xallyzach
Wallet shares move far less dramatically than volume shares around fee dates, but still maintained a similar cohort pattern, where the three markets mentioned above showed the most churn in the Informed/Pro user cohort.
Pantera Research Lab@sui414 · @0xallyzach
Entry Timing by Cohorts
Within a market’s life, when do the tribes act? Late entry is the norm — how late depends on the market’s information game. Sports and Crypto are the most latency-competitive: every cohort’s median entry sits at 92–95% of the market’s life — adversarial selection pressure for retail. Culture follows at 85–90%. Weather is the opposite extreme: entries come earliest of any category (retail median 61%, bots 51%, on a small book of 15 qualifying markets), consistent with outcomes that predictive models can price well before resolution — and it’s the one category where pros wait longest (83%) relative to retail.
Among our datasets, only markets whose complete life is observed qualify: resolved within H1, with fills traced back to each market’s first trade — including pre-2026 entries.
Pantera Research Lab@sui414 · @0xallyzach
The chart below shows the daily cohort volume composition of each category’s largest market by H1 volume. Retail action often runs throughout the market’s duration and across all categories, while Informed/Pro and bot actions are more likely to spike as information comes in, or accumulate at resolution time near the end.
Pantera Research Lab@sui414 · @0xallyzach
User PnL: who actually makes money?
Add up every dollar that flowed through every wallet’s positions in markets resolved during H1 and the ledger closes at zero — before fees: 34% of the 1.97M wallets with resolved positions ended positive. The tails balance almost perfectly: +$666M big winners vs −$530M big losers; the total, including all groups, nets to ~$0, confirming this is gross-of-fees PnL (trading is zero-sum before the house takes its cut — subtracting the ~$184M of H1 fees would push the aggregate negative).
Pantera Research Lab@sui414 · @0xallyzach
Split the ledger by tribe and the transfer has a direction: retail and pros fund the bot cohort’s gains.
Bots are the only net-positive cohort in terms of total PnL (+$108M) — yet only 27% of bot wallets profit. The median bot wallet loses money. The class’s profit is carried by a small number of large operations, and persistence is vanishingly rare: of 1.75 million wallets, 146 were profitable in all six months.
Pros have the best hit rate: 42% of wallets profitable, compared to the 35% of retail and 27% of bots — but they total to a loss of −$26M. Note that given the various incentive programs and rebates not accounted for in this gross PnL accounting, this does not necessarily mean pros are net losing money.
Retail funds the pool (−$74M). 303,837 wallets with resolved positions but no H1 trading (positions carried in from 2025; −$7.2M, median ≈ $0) are excluded from the cohort view below. Our heuristic-produced cohorts separate style and aggregate economics cleanly (bots extract, retail pays), but within every cohort the dispersion is huge — a handful of large bots carry the class while most individual bot wallets lose.
Pantera Research Lab@sui414 · @0xallyzach
Pantera Research Lab@sui414 · @0xallyzach
Among the 1.75-million-wallet user base, 146 wallets were profitable all six months. Among wallets with any profitable month, a large share made >90% of their profits in a single month (amber).
Pantera Research Lab@sui414 · @0xallyzach
How PnL is measured
Realized PnL per wallet and market, for markets resolved within H1: signed cash flows (buys and splits out; sells, merges and redemptions in) plus terminal settlement of residual shares at resolution — each market a closed loop, no cost-basis accounting needed. Figures are gross of fees; the tiers netting to ~$0 across two million wallets is the accounting identity closing, and subtracting the ~$184M of H1 fees pushes the aggregate negative. The approach follows Sergeenkov’s cash-flow method with the terminal-settlement refinement of Dai, Jia & Yu (arXiv:2606.31675). Addresses are not people: multi-wallet operations are invisible to any per-address accounting.
Acknowledgment
Our thanks to the Surf team for providing the MCP tool and datasets for the research. The user onboarding analysis framework draws on @0xtaetaehoho’s “Listings are the Product” article.