Unitree on Hyperliquid: 3 signals to read a pre-IPO perpetual

Unitree closed its first day of trading at 845 yuan on August 19, 2026 —460% above its IPO price— and of the two mechanisms that had priced it before the debut, the order book for its perpetual on Hyperliquid, with $9.1 million in live positions, fell between 20% and 26% short; the IPO underwriting syndicate, 82%. The book —the record where buy and sell orders for a contract meet— erred between 3.2 and 4.1 times less than the banks that set the offering price and distributed the shares, and both failed in the same direction twelve days apart. With Unitree, there are now four public episodes of equity perpetuals on Hyperliquid (futures without expiry, here based on the valuation of a company not yet listed): three pre-IPO resolutions —Cerebras, SpaceX, and Unitree— and the transition to external pricing for SK Hynix. This article reconstructs Unitree's eight prices against a share count verified through four channels and outlines three verifiable signals before the next one resolves.

How much was Unitree worth according to its perpetual vs. its own IPO?

Unitree Robotics manufactures quadruped and humanoid robots in Hangzhou. In 2025, it shipped over 5,500 humanoid units, generated 1.699 billion yuan in revenue, and reported a gross margin of 60.13% in its core business, according to the prospectus cited by Robotopian. It is listed on the STAR Market —the technology segment of the Shanghai Stock Exchange— under the ticker 688836.

Between August 4 and 19, 2026, eight valuation figures for the same company circulated: two from the regulated placement process, three from the 24-hour open order book on Hyperliquid, and three from the spot market on the day of the debut.

The last two perpetual figures do not match, and we do not choose between them here: CoinDesk, using Allium data, places the contract between $92 and $94 on Friday the 14th; Bloomberg, cited by Cryptonomist, puts it around $100 on the 18th. The scorecard is calculated using both.

The book that produced these prices was small: in the August 14-15 snapshot, it totaled $9.1 million in open interest —the value of open positions not yet closed— and about $59 million in volume across two independent markets, one from Trade.xyz and another from Paragon, the N8V Labs deployer led by Taha El-Magbri, which launched markets on Hyperliquid on April 2, 2026. When both were active, they traded with an average difference of 1.6%, according to CoinDesk, which by the morning of the debut already counted about $29 million in open interest. On the other side, an IPO of approximately $904 million with retail requests roughly 8,000 times the offered tranche.

Who came closer to the real price: the order book or the underwriters?

The comparison requires choosing a benchmark, and on August 19, three valuations from the same day circulated that seemed contradictory. $66 billion is the first trade, the opening at 1,100 yuan (CoinDesk, Fortune). $53 billion and +487% is the morning session close, at about 884 yuan, which was reported as the "close" in the U.S. (Crowdfund Insider, CoinDesk). $50.7 billion and +460% is the full session close, at 845 yuan (MEXC, Reuters via Yahoo). All three are correct for different moments of the same day.

This analysis measures against the first-day close of 845 yuan. The STAR opening is an auction price without a fluctuation band —the segment does not apply one during the first five sessions— so the first trade captures all the pent-up demand from a subscription oversubscribed 8,000 times; the close is the only one of the two that survived an entire session.

With that denominator, the scorecard looks like this: the IPO price was 82.2% below the close; the perpetual was between 20.2% (using the 18th price) and 25.8% (using the 14th price) below. The $9.1 million book erred between 3.2 and 4.1 times less than the $9 billion placement. Against the opening, the point that favors the traditional process, the IPO price errs by 86.3% and the perpetual between 38.7% and 43.0%: the book still errs about twice as little. The ranking remains the same with either moment; only the magnitude changes.

That denominator also determines how the sector's own track record is read. In the filing that the Hyperliquid Policy Center and Trade.xyz submitted to the SEC on August 18, 2026 —the legal side is in our breakdown of the five Washington catalysts— the operator provides five resolved contracts whose final price was, in the best case, within 0.44% of the underlying's stock market opening and, in the worst, within 7.23%. Unitree hits 43% by that measure: six times above the bad end of the range the deployer presented as its track record on the eve of the debut.

How many shares does Unitree have and what is it worth at each price?

The valuations over these three weeks seemed incompatible —4.2 billion, 42 billion, and 205 billion yuan for the same company— but they are not: they all result from multiplying a share price by the same number of shares, and that number is public.

All eight rows fit within the source's rounding, and the count is confirmed by four independent routes: the Shanghai Stock Exchange note, the arithmetic of the 40.44 million shares placed, the 342 billion yuan that MEXC published for the close —845 × 404.4 gives 341,718— and the +487% from the morning session, which only squares by multiplying 884 yuan by that same capital.

Two consequences follow. The first resolves the conflict of the yuan figures: 42 billion and 205 billion RMB are both correct for different moments —the prospectus and the August 4 perpetual— and the 4.2 billion that also circulated is a factor-of-ten error. The second affects the multiple that dominated headlines: the "4.88x potential" at launch was calculated against the prospectus valuation ($74 vs $15.41), because on August 4, no IPO price existed yet. Against the actual price, the multiple was 3.3 times.

Why is no arbitrage possible in a pre-IPO perpetual until the debut day?

A standard perpetual has an external anchor: the mark price —used to calculate liquidations and funding— comes from an oracle that averages reference markets. Bitcoin trades in fifty places and the oracle takes the average; a private company trades in none. Under HIP-3, the framework that allows any team with sufficient HYPE —the protocol's token— in collateral to deploy their own market on Hyperliquid and operate their own oracle, the deployer of a pre-IPO does what would be an anomaly in another derivative: the mark price comes from the order book itself, because there is no external source to read. The product construction is in our piece on Trade.xyz and the economic distribution of the deployment in the one on builder codes and HIP-3.

From this comes the property that dictates everything else: while the market is its own oracle, no arbitrage is possible. No one can buy the underlying cheap and sell the contract dear, because the underlying cannot be bought; the price does not converge toward anything, it reflects what the two sides of the book cross. On the day of the debut, this shuts off instantly and the contract imports the price from the market where the stock is now trading. As of August 23, 2026, neither Trade.xyz nor Paragon has published the exact date the UNITREE contract switched to external pricing; the dashboard already shows different mark and oracle prices, a sign that the external oracle is active.

Regarding the size of the layer where all this lives, published figures are not comparable and should be stated with dates. FinanceFeeds reported over $200 billion in cumulative volume as of July 1, 2026, and a peak of $3.2 billion in open interest in June. In August, Cryptopolitan and Cryptotimes documented the first close above $4 billion in open interest —4,030 on August 8— and the Loris Tools dashboard estimated 3,870 on the 20th. The over 460 billion attributed to Trade.xyz since January (KuCoin) measures something else: volume from a single deployer, not the layer's open interest. Three metrics that should not be chained together.

Why was the Cerebras error 1.3% while Unitree's reached 26%?

Four cases, four results that look nothing alike, with a reading warning in the "Reference" column: the Cerebras scorecard was published against the opening price, while SpaceX and Unitree are measured here against the first-day close.

Book depth does not rank the cases. The SpaceX contract reached the eve of its IPO with about $216 million in open interest on an implicit valuation of $1.78 trillion —million millions—: 0.012% of the company live in the book, compared to Unitree's 0.024%. The one that was thinner in proportion erred by seven points; the one with double the relative thickness erred between twenty and twenty-six. Depths of the same order, errors separated by a factor of three.

Proximity to the debut doesn't explain it either. The SpaceX post-mortem left distance to the trade as a rule of thumb —the closer the price is taken, the smaller the error— and Unitree breaks it: its price was taken on the eve, closer to the debut than SpaceX's, and erred three times more. What replaces that rule is where the stock was going to list. Cerebras and SpaceX debuted on Nasdaq, within reach of anyone trading the perpetual: being able to buy or sell the underlying on Day D disciplines the contract weeks in advance. Unitree debuted on the STAR, an A-share segment where foreign access goes through QFII —the Qualified Foreign Institutional Investor regime— and Stock Connect, and is reserved here for institutions. Those trading the perpetual on August 18 were not going to be able to buy the stock on the 19th, nor short it, nor hedge: that book wasn't estimating a price, it was aggregating opinions on a price it couldn't touch. And with the subscription oversubscribed ~8,000 times, almost all retail demand was left out of the allocation and hit the secondary market on the first day with unallocated cash: a buying pressure that a synthetic dollar book cannot measure because it does not participate in it.

What happens to the contract the day the company starts trading?

The transition to external pricing is where these contracts change nature and where they have already failed. On July 27, 2026, a single share of SK Hynix was traded on the Nextrade pre-market, the alternative Korean exchange, at the daily downward limit; the oracle brought that trade to the SKHX perpetual mark price, and about 960 long accounts were liquidated with about $57 million in notional, according to the minute-by-minute reconstruction of the SK Hynix flash crash.

For reading pre-IPOs, the lesson is one of framing: that was not a wrong forecast, but a failure of the conduit bringing the price from outside. While the contract is marked against its own book, neither the conduit nor the risk exists; the day after the debut, both exist. And the STAR, with a ±20% fluctuation band starting from the sixth session, meets the conditions of the Korean episode: a reference price that can jump suddenly and a synthetic book forced to follow it.

How to distinguish a pre-IPO price with information from one that just aggregates bets?

Bybit already lists pre-IPO perpetuals for Unitree and Moonshot AI —the Chinese developer of Kimi models— so the format has left Hyperliquid and will appear more frequently and with less context. These are the three possible checks before the contract resolves, with what each says and what it doesn't.

The third requires a warning. Two markets trading at a 1.6% distance look like cross-validation but are something else: they are operated by highly overlapping populations, with the same public information and no way to contrast it against the asset. And the layer is concentrated: public counts attribute over 90% of HIP-3 open interest to Trade.xyz (FinanceFeeds, July 1) and 99.5% on August 8 (Cryptopolitan); neither was taken on the same day as the Unitree book.

The checking order is as follows:

1. Look at where the stock will debut and if the contract trader will be able to buy or sell it there on Day D. Nasdaq and the STAR do not produce the same error.
2. Look at open interest and volume in proportion to the implicit valuation, not in absolute value: 9.1 million over 38,000 is 0.024%.
3. Look at how many deployers have a market on the same underlying, knowing that agreement between books measures consensus, not accuracy.
4. Look at the scorecard denominator. Accuracy percentages are presented against the stock market opening price; measuring against the first-day close or the thirty-day price changes the result by a factor of two, as Unitree shows (43% vs 26%).

What would have to happen for the bearish bias of pre-IPO perpetuals to be false?

With three resolutions of comparable price, the sign of the error is not stable: Cerebras and SpaceX stayed above the result (1.3% and ~7%) and Unitree far below (between 20.2% and 25.8%). What ranks the three cases is the magnitude, and the variable that ranks it is whether the underlying was going to be accessible to the contract trader. The hypothesis is that the bias becomes bearish when the underlying is not arbitrable from the perpetual market, because without the possibility of hedging, the book stops incorporating the demand that only exists in the local market.

Falsification Test. If in the next two resolutions the perpetual price stays above the underlying price thirty days after the debut, the downward bias thesis falls and must be rewritten as a symmetric bias. If this also happens in a debut outside Nasdaq, the arbitrability explanation also falls. Both conditions are verifiable with public prices and will be reviewed here when they resolve.

Meanwhile, the data from August 19 is that a $9.1 million book came between 3.2 and 4.1 times closer to Unitree's market value than the process that placed its shares, and yet it still fell a fifth short in the best of cuts. Both halves of the sentence are the same information: these prices contain signal, and the amount depends on conditions verifiable in advance, not on the enthusiasm with which they are published. The question left open by the SpaceX post-mortem —whether the on-chain market was right or just lucky— now has a fourth data point, and it says neither: it is right where it can be corrected.

Sources and links: CNBC — IPO price of 150.80 yuan · CoinDesk — debut outpaces perpetuals (Allium data) · Fortune — +460% and 66 billion at opening · MEXC — close at 845 yuan and 342 billion RMB · Gasgoo — 40.44 M shares and oversubscription · Shanghai Stock Exchange — enlarged capital of 404.5 M shares · KuCoin — perpetual launch at ~$74 · Cryptonomist — perpetual at ~$100 (Bloomberg data) · FinanceFeeds (July 1) — HIP-3 aggregates · Cryptopolitan — HIP-3 above 4 billion · Loris Tools — dashboard by deployer · PR Newswire — Paragon deploys on Hyperliquid · Robotopian — Unitree 2025 accounts