Hyperliquid’s HIP-4 prediction market may be one of the most tightly integrated products in crypto trading, but scale remains its central problem, according to a report by IOSG Ventures author Mario Chow. Using data through Oct. 5, 2026, the report says HIP-4 has generated $317 million in cumulative volume since launching in May. Kalshi, by comparison, can do that in less than four hours. In September, for every $1,000 traded on Kalshi and Polymarket, HIP-4 handled about $0.70.
The report’s core argument is that HIP-4 does not really resemble Polymarket as much as it resembles an option sitting next to a perpetual position. A Yes share and a leveraged position live in the same account, draw from the same margin, and trade on the same matching engine. That keeps a directional bet and the hedge for that bet inside one interface. Three other features also set it apart: anyone staking 500,000 HYPE can open a market; settlement conditions are fixed in a template before the first trade; and users pay nothing to enter, then roughly 0.1% to exit, the reverse of how Kalshi and Polymarket charge.
For the first four months, Hyperliquid was effectively the only operator. When HIP-4 launched on May 2, 2026, all 691 markets were run directly by the protocol. External venues were not allowed until Aug. 29. Since then, three operators have posted the required stake, and there is still no fourth.
Visibility depends on where a user is located. Prediction markets have a dedicated page at app.hyperliquid.xyz/outcomes, and the frontend decides by region whether to place that section alongside perpetuals and spot in the main trading list. The report says a connection from South Korea showed the category in the market list, while the network used in the authors’ test did not show it in the list or menu, even though the page itself remained accessible and tradable. Hyperliquid restricts frontend access in several countries, so the entry point a user sees can differ across regions.
Four product differences versus Kalshi and Polymarket
The report reconstructs HIP-4’s fee schedule from trades where fees were actually charged. It says the base rate is exactly double Hyperliquid spot fees: spot charges 0.070% for takers and 0.040% for makers at the base tier. In HIP-4, one side is doubled and the other is reduced to zero, leaving the round-trip cost equal to spot, except that the full charge is collected on exit.
Discounts stack through simple multiplication. A 4% referral discount cuts 0.140% to 0.1344%, and a further 10% staking discount brings that to 0.12096%. The author says every checked trade matched that math exactly. Makers do not receive rebates, which the report describes as the only place where market makers are worse off here than in other parts of the exchange.
No venue has yet raised its own fee multiplier above the default, so venue revenue is carved out of the protocol fee rather than added on top. The only possible charge on entry comes from frontend apps. Under spot rules, builder fees can go as high as 1%, and one app was charging 0.81% on entry, roughly six times the protocol’s exit fee.
Settlement is where venue risk becomes more visible. Polymarket relies on UMA tokenholder voting, and the report points to a $375 million Microstrategy market earlier this year that became highly contentious. Kalshi, as a regulated exchange, settles markets itself under a regulatory framework. HIP-4 splits the process in two. Crypto price markets settle automatically by reading the exchange mark price at the specified minute. All other markets are settled by the market creator signing the result. There is no dispute window, and no one else can sign in that party’s place.
Of the 228 markets open at the time of the report, 223 fell into that second category. The constraints are financial and temporal rather than procedural. As long as a venue still has unsettled markets, its stake remains locked. The stake cannot be withdrawn for six months. Market wording must also come from validator-approved templates. Those templates are intentionally long and exacting. The report notes that the template for a company listing spends an entire page defining what counts as a listing. Writing the rules before trading starts is better than arguing after the fact, the author says, but handing the pen to a named party without an appeal path is still weaker than a functioning procedure, assuming that procedure itself works.
World Cup volume drove the peak, then daily markets faded
June was the high point, and the World Cup was the reason. Football markets alone generated $89.5 million, or 28% of HIP-4’s cumulative volume. Once the tournament ended, volume fell by more than 80% over the next two months. September recovered to $51 million, but all of that rebound came from external venues. The first five days of October produced $5 million.
Underneath those event spikes, the protocol had been running the same market every day since May: whether Bitcoin would be above a given level at 6 a.m. the next morning. The report calls it perhaps the most Hyperliquid-native product in the whole set, and says it fell for four straight months before bottoming. In the week of May 11, that series traded $16.6 million. By the week of Sept. 28, it was down to $1.2 million, with the prior two weeks at roughly the same level. The conclusion is blunt: ask the same question every day and eventually almost nobody wants to answer it.
Five weeks after opening deployment, incentives are carrying the rebound
Outcome was the first external venue to go live on Aug. 29. Skew registered markets the next day and has seen almost no trading since. Events, operated by Trade.xyz, registered on Sept. 5 and opened to the public on Sept. 10.
Looking at the most recent day across all HIP-4 activity, Outcome accounted for 68% of volume, Trade.xyz for 21%, protocol-run markets for 10%, and the remainder for Skew. Restrict the view to external venues only, and Outcome has taken 92% of volume since Aug. 29, while Trade.xyz has climbed from zero to nearly one-quarter on the latest day. The report says Trade.xyz has been more opinionated in market selection, listing binary contracts on its own stock and commodity perpetual prices, NFL markets, and the only exchange-listed questions on whether Anthropic and OpenAI will go public before a deadline.
That volume comes with an asterisk. Outcome is running a $1 million rewards program that pays makers and traders by market and by day, but only for orders carrying its own app code. As of Oct. 5, it had distributed $273,409 to 2,487 wallets, or about $7,400 a day. The report says that equals roughly 0.6% of Outcome’s total traded volume, about four times the cost of a round trip. One-quarter of the budget has already been spent, and daily payouts have not slowed.
External venues were doing roughly $1 million to $2 million a day in September, with weekends the busiest. In October, that slipped back to around $1 million a day, while rewards kept flowing.
Open deployment has also produced the first duplicate markets. Outcome and Trade.xyz both listed a market on the Federal Reserve’s October rate decision. Over the past seven days, Outcome’s version traded $354,000. Trade.xyz’s version traded $90. Both venues also listed a set of questions on Anthropic’s first-day public market capitalization, and neither side saw much activity. The report’s point is simple: if anyone can list a question, two venues can list the same question, and their order books do not merge. Flow goes to the venue paying for it.
Trading mix is tilting toward sports. Over the past 30 days, football, American football, and esports made up 45% of volume. Crypto price questions accounted for 42%, stocks and commodities 7%, and Fed-related markets 6%. Open interest is even more concentrated. About two-thirds of capital at risk sits in sports, mostly season-long books such as the Premier League, Champions League, and NFL title markets. Crypto price questions account for 17%.
Template growth is outpacing venue growth. Validators have approved 28 templates, up from 18 in mid-September. New additions include point spreads and totals, the Ballon d’Or, head-to-head contests between two AI models on a leaderboard, and first-day market capitalization for newly listed companies. The next network upgrade will double the cap per venue, taking simultaneously live markets to 200 and daily new listings to 1,000. Testnet has 95 registered deployers. Mainnet still has three.
Where the orders come from: about 70% of flow runs through Outcome
Frontend apps that place orders on a user’s behalf require authorization for a separate signing key, and that key is named after the app. A wallet that registered through outcome.xyz therefore keeps a public label such as Outcome-9d3c74de. Hyperliquid’s own mobile login leaves a similar trace, shown as Mobile QR. Desktop trading does not. Each action is signed directly with the user’s own key and leaves no app label.
The report only studies Outcome because it represents roughly 90% of external-venue volume. It finds that about 70% of trading came from wallets that had registered on outcome.xyz, roughly one-fifth came from wallets with no label at all, and about one-tenth came from scripts and third-party tools using self-named keys. The larger the wallet, the larger Outcome’s share. Among the biggest 20 wallets, Outcome represented three-quarters of traded value. Across the rest, it was about half. That is exactly what the rewards program should produce, the report says, because only orders routed through Outcome’s code qualify for rewards, and the market makers collecting those rewards are also the largest wallets.
The method has limits. A label only proves that a wallet once registered through an app, not that any specific trade was routed by it. The unlabeled fifth cannot be treated as exchange-native users either, because most of those wallets were quoting both sides across many order books with scripts signed by their own keys. The sample covered 121 wallets drawn from recent trades in the 60 most active order books and weighted by 48-hour traded value, which naturally biases the sample toward market makers. Fees do not fully solve the attribution problem. Less than 1% of sampled value paid builder fees, totaling just $9 over two days, and Outcome’s own code charges nothing, making free routing look the same as no app at all.
The gap with Kalshi and Polymarket is still widening
The report puts the scale gap at roughly 1,400x, and says it is still widening in dollar terms even though HIP-4 itself is growing. Kalshi traded $59.3 billion in September, up $20.6 billion from August. Polymarket, including its U.S. business, traded $13 billion, with most activity now concentrated in the U.S. app. HIP-4 did $51 million.
Legacy platforms are not winning on market design, the report argues. They are winning on the two things that actually drive volume. First, they have events people care about, in the thousands, and can list them within hours of a news break. Second, they have distribution: apps, brand, sports audiences, and in Kalshi’s case a compliant wrapper that U.S. brokerages can connect to directly. HIP-4 has 28 templates, three venues, and a user base that was already trading perpetuals.
Those incumbents are also moving toward Hyperliquid’s model. In early September, Polymarket launched perpetual futures for users outside the U.S., with 67 markets across crypto, stocks, indexes, and commodities, and leverage up to 20x. Keeping a prediction bet and a leveraged position under one roof used to be HIP-4’s distinctive feature. Now it is part of a feature race, and Polymarket entered that race with event supply already in hand.
The economics do not yet work
At the pace of the past 30 days, HIP-4 traders would pay about $480,000 in fees over a full year. Venues can take at most half of that, and three of them have to split the pool. By contrast, the stake required for one venue to enter the market would earn $1 million a year if it simply sat idle, and the largest venue by volume is paying rewards at an annualized rate of $2.7 million to buy that flow.
On those numbers, the report says no venue is really operating a business yet. They are buying an option. That option has value: Hyperliquid has said deployers may later be able to raise fees themselves; volume could eventually grow enough to support costs; and private-market valuation comps in the sector stand at $22 billion and $21 billion, while Kalshi is discussing a $40 billion valuation. But it is still an option, and the carrying cost is a $47 million stake. If the volume never arrives, the option expires worthless.
Better design, but the market is not here yet
The report ends on a restrained conclusion. Hyperliquid has assembled the better product on the dimensions engineers tend to care about: one account, one margin pool, prediction positions and hedges on the same screen, settlement terms written into templates before the first trade, and a shared order book for Yes and No so liquidity is not split in two. Across the past five months, the data show no sign that design itself is the bottleneck.
The strongest case for HIP-4 is that users never have to leave Hyperliquid. Margin, hedging, and prediction positions all sit in one account. For a desk already trading perpetuals on Hyperliquid, adding a prediction position is one extra click, not a platform switch, a new balance, or a transfer back and forth. The report says that remains valuable regardless of how competition with Kalshi plays out, and it is the reason HIP-4 could persist at some scale even if it never becomes a large standalone business. Still, it is less unique than it was in August. Polymarket now sells leverage next to prediction bets too, and it started from the side that already had users.
Depth is still thin. Across the 25 most active order books, the median bid-ask spread is 0.29 cents, and about $4,700 sits within one cent of the best ask. A $1,000 market buy only clears a few basis points above the best ask. A $10,000 buy pays 2% to 3% more, and in four or five of those 25 books, visible liquidity is not enough to fill the order at all. Retail-sized trading works. Larger size does not, at least not cleanly. Market makers constantly refresh quotes, so patient limit orders can do better than those snapshots suggest, but the books still look like what they are: a five-month-old market.
There is also not much to trade. Prediction-market volume is built from topics, and topics usually come from editorial teams listing hundreds of events each week or from sports calendars. HIP-4 has 28 approved phrasings. Three venues have each posted a $47 million stake to use them, and two of those venues are already listing the same questions. Kalshi can open a market around a news event in a single afternoon.
The user base is mismatched in a very specific way. Each address trades eight times a day, according to the report. That is a trading desk, not a mass-market bettor. If the goal is to give crypto traders a capped-payout instrument, that is fine. In fact, that is what the design does best. If the goal is to win the audience Kalshi and Polymarket are chasing, it is not. The fact that two-thirds of capital at risk sits in season-long football books suggests venues are still pushing in that direction anyway.
The most likely outcome, the report says, is the unglamorous one. HIP-4 remains a feature inside Hyperliquid, earns a small amount, and matters because perpetual traders do not need to open accounts elsewhere. That is worth building. It is not enough, at least for now, to value it as a standalone prediction-market business.