Source: https://tesseralytics.dev/blog/hyperliquid-vs-perp-dexes

[← Blog](https://tesseralytics.dev/blog)

Comparison 17 November 2026

# Hyperliquid vs. the perpetual DEX field (2026)

Fifteen perpetual DEXes still report volume, and two that most comparison tables list are dead, so this roster is a probe with a date on it. Side by side on fees, listings, funding and chain, plus what the published fees hide: the 2.86 bps actually paid on Hyperliquid’s tape, and why published schedules cannot answer “cheapest”.

Roster probed 17 November 2026 at 13:16 UTC. Fees and depth measured over 2026-10.

Every comparison of perpetual DEXes comes down to three questions: which one is cheapest, which lists the most markets, and which has the deepest books. The first has no answer in any published fee schedule, and this post is mostly an argument for why. The other two have answers, they just are not the ones the marketing pages give.

The field is also smaller than it looks. Of the venues that turn up in every "Hyperliquid vs" table, two are dead: Vertex shut down in July 2025 with its deployments wound down and its successor moved to Ink, and Zeta stopped trading on 1 May 2025. Drift no longer exists under that name; it relaunched as Velocity in July 2026 with five markets. And one survivor changed chains in October 2026: Extended moved from Starknet to Circle's Arc on 22 October, twenty-six days before this roster was built. **A comparison table is a dated artifact**: every number below was true on the day it was checked, and the field moves faster than the posts that compare it.

What does get decided by the numbers: Hyperliquid is 57% of the notional across the 15 venues that report any, and it trades 378× what dYdX does, which is the comparison this question was originally about. It is not the most listings, and it is not the deepest per market once you leave bitcoin and ether. Its real advantage is the one nobody publishes: it is the only venue in this field where the fee actually paid can be measured from public data, and the measurement says 2.859 bps all in over 2026-10 (299 coins, $243.06b of notional), against the 4.5 bps on its fee page.

The short answer

Pick the row that matches your question.

The most markets to trade

Aster (641, all trading)

The most notional in one book

Hyperliquid (57% of the field)

The deepest majors

Hyperliquid, by a wide margin

Zero published fees

Variational, Lighter, Paradex

A measured all-in cost

Hyperliquid only, so far

## The field, side by side

| Venue | Settles on | Perps (trading) | 24h notional | Maker | Taker | Funding |
| --- | --- | --- | --- | --- | --- | --- |
| Hyperliquid | its own L1 (HyperCore) | 565 (350) | $10.04b | 1.5 → 0 | 4.5 → 2.4 | 1h |
| Variational (Omni) aggregator only | Arbitrum | 580 | $2.03b | 0 | 0 | 1–8h |
| Aster | multichain | 641 (641) | $1.91b | 0 | 4.0 | 8h |
| Lighter | Ethereum L2 | 245 (202) | $1.84b | 0 | 0 | 1h |
| edgeX aggregator only | not verified | 137 | $453.0m | 1.8 → 0 | 3.8 → 2.4 | n/a |
| Nado | Ink L2 (57073) | 83 | $401.1m | 1.0 → −0.8 | 3.5 → 1.5 | 1h |
| ApeX Omni aggregator only | not verified | 128 | $381.1m | n/a | n/a | n/a |
| Extended | Arc (Circle L1) since 2026-10-22 | 416 (242) | $289.3m | 0 | 2.5 | 1h |
| RISEx aggregator only | not verified | 53 | $96.6m | n/a | n/a | n/a |
| GMX (Arbitrum) | Arbitrum / Avalanche / Solana | 131 | $48.9m | n/a | 4 or 6 | continuous |
| Orderly | Orderly chain (shared book) | 149 | $38.1m | 0 → −0.2 | ≥3.0 → 1.0 | 4h |
| dYdX v4 | dYdX Chain (Cosmos appchain) | 296 (43) | $26.6m | 1.0 → −1.1 | 5.0 → 2.5 | 1h |
| Aevo | OP Stack L2 (off-chain matching) | 102 (102) | $3.9m | 3.5 | 6.5 | 1h |
| Paradex | Paradex Chain (Starknet appchain) | 65 (38) | $1.9m | 0 | 0 | 8h |
| Velocity (ex-Drift) | Solana | 5 (5) | $676k | +0.25 paid | 4.0 → 2.0 | 1h |

Notional is the venue's own 24h figure where it publishes one, otherwise CoinGecko's, and the row says which. "Aggregator only" means no public market endpoint was found, so both the market count and the notional are CoinGecko's; on the other rows the count is the venue's own and only an absent notional falls back to the aggregator. Counts are markets the venue lists, with those that traded in the last 24 hours in brackets; the two differ a lot at Hyperliquid and barely at all at Aster. Fee columns are the published tier-0 schedule, best reachable tier after the arrow, before any staking discount.

Two rows are missing from that table because they no longer trade: Vertex (shut down July 2025; assets moved to Nado on Ink) and Zeta (ceased trading 1 May 2025). Both still appear in comparison tables written before those dates.

## Five venues are the field

Sort by notional and the field collapses fast. Hyperliquid trades $10.04b a day. The next four (Variational (Omni), Aster, Lighter, edgeX) are all between $453.0m and $2.03b. Everything else in this table is small against them, and the 2024 cohort, meaning the venues this comparison used to be about, is the smallest part of it: dYdX v4 at $26.6m, Aevo at $3.9m, Paradex at $1.9m.

That reordering is the finding, and it is not a compliment to anybody. It means the field has one venue with a real book and several venues with a real product but no book, and in practice traders now ask which chain holds their collateral while they trade the same perps, not which DEX they trade on.

It also means listing counts mislead in both directions. 641 Aster markets sounds like a wider universe than 565 Hyperliquid ones, and it is, but Hyperliquid's total is 235 core perps plus 330 on ten builder-run venues, of which only three cleared $1m in a day. Of Aster's 641, every single one traded. Read the second number in brackets, not the first.

## Published fees rank pricing models

Here is the table every comparison post has, and here is what it actually ranks. In this field three venues publish a zero taker fee for a normal account: Lighter, Paradex and Variational (Omni). One more, Orderly, sells through builders who set the trader's fee on top of a base rate, so a trader there can see zero as well.

None of those venues is giving away execution. Variational says it plainly in its own documentation: there is no trading fee because the Omni LP's revenue is the quoted spread, and a liquidation carries a ±0.5% penalty. Lighter's standard accounts pay nothing while Plus and Premium tiers pay 0.5 and 2.8 bps of taker fee, which is to say the product is priced for someone, just not for the account in the fee table. Orderly's base fee is paid by the builder, and the builder recovers it.

So sorting this field by its published fees produces a ranking of pricing models, not of costs. The cost a trader actually pays is fee plus spread plus impact plus funding, and of those four only the fee has a number on a website.

**On one venue we can measure the real one.** Tessera's tape carries `total_fees`, `taker_fees` and `maker_fees` per minute against a notional reconstruction, so the fee Hyperliquid actually charged is an identity, not an estimate.

| window | coins | notional | all-in fee | of which taker | of which maker |
| --- | --- | --- | --- | --- | --- |
| 2026-10 | 299 | $243.06b | **2.859** bps | 2.782 | 0.077 |
| 2026-09 | 287 | $247.29b | **2.986** bps | 2.915 | 0.071 |

All-in fee is total fees divided by notional, volume-weighted over every coin that traded in the month. The two windows are consecutive and agree to 0.13 bps (2.859 and 2.986); what separates them is month mix, not a price change. The identity `taker + maker = total` holds on every coin-month, to within a hundredth of a cent across the whole tape.

2.859 bps against a published 4.5 is 64% of the number on the fee page. The gap is not a discount: it is volume tiers, HYPE staking, maker rebates and the venue's own HIP-3 fee schedule, and 97.3% of the fee dollars came from takers. Anyone quoting 4.5 bps for Hyperliquid is quoting a ceiling that almost nobody pays.

## The split that explains the rest

Hyperliquid is two fee regimes under one orderbook. Core crypto perps charge the schedule everyone quotes. HIP-3 builder perps, the tokens like `xyz:NVDA` that a third party deploys and prices itself, run at a fraction of it because growth mode cuts the rate by around 90% while the deployer takes a share.

| class | coins | notional | share of tape | all-in fee | median depth |
| --- | --- | --- | --- | --- | --- |
| Core crypto perps | 179 | $167.38b | 68.9% | 3.676 bps | $1,268 |
| HIP-3 builder perps | 120 | $75.68b | 31.1% | 1.050 bps | $7,978 |

Median depth is the dollars of takedown that move the price one basis point, per coin, so the two columns are not additive. HIP-3 is 120 markets carrying 31% of the tape at 3.5× lower fee than the core book.

This is why a single fee number for "Hyperliquid" is meaningless, and why the aggregate moved between the two months only because the mix did. It is also the competitive fact the comparison tables miss: the cheapest perps in this field are on Hyperliquid's HIP-3 book, at 1.05 bps all in.

## The tail does not pay more

The folk claim is that thin markets are expensive and liquid ones are cheap, so a fee table should show a gradient. Sort all 299 coins into ten equal groups by notional and the volume-weighted rate barely moves: 2.14 bps in the smallest decile, 3.10 in the largest. Fee tiers are a trader-level construct and trader mix does not vary enough by coin for that to show up at coin level.

What does vary by coin, and by four orders of magnitude, is depth: $1,456,452 of takedown moves the deepest coin that clears the floor one basis point, $99 the thinnest. That is the real cost gradient, and none of the published fees above reflect it.

| decile (1 = smallest) | coins | notional | fee, volume-weighted | fee, median coin | median depth |
| --- | --- | --- | --- | --- | --- |
| 1 | 29 | $63.1m | 2.14 | 2.21 | $264 |
| 2 | 30 | $155.4m | 2.78 | 2.71 | $331 |
| 3 | 30 | $252.0m | 2.84 | 2.90 | $527 |
| 4 | 30 | $447.4m | 3.01 | 3.31 | $1,037 |
| 5 | 30 | $856.5m | 2.78 | 3.24 | $1,353 |
| 6 | 30 | $1.57b | 3.07 | 3.75 | $2,296 |
| 7 | 30 | $2.86b | 2.90 | 3.71 | $4,889 |
| 8 | 30 | $6.15b | 3.10 | 3.73 | $5,628 |
| 9 | 30 | $17.23b | 2.50 | 2.38 | $12,704 |
| 10 | 30 | $213.48b | 2.88 | 1.10 | $79,322 |

The top decile carries 88% of the month's notional, so its volume-weighted rate is the venue's rate and its median coin (which is mostly HIP-3) is not. All figures in basis points except depth, which is dollars.

## Depth, measured per coin

Depth is measured as Kyle's lambda: regress the minute return on the minute's signed aggressive notional, and invert it into the dollars of takedown that move the price one basis point. It is a lower bound, because one-minute aggregation smooths within-minute impact away, but it is computed identically for every coin, which is what a comparison needs.

Across the 259 coins that clear a $5m month, that number spans $1,456,452 down to $99, with a median of $3,433. Bitcoin and ether sit at the top of that range, hundreds of times the median coin.

| market | notional | taker fee | to move 1 bp | $1m round trip | R² |
| --- | --- | --- | --- | --- | --- |
| BTC | $72.82b | 3.72 | $1,456,452 | 8.8 bps | 0.197 |
| ETH | $31.44b | 3.27 | $741,139 | 9.2 bps | 0.137 |
| HYPE | $15.02b | 3.29 | $103,574 | 25.9 bps | 0.276 |
| ZEC | $10.13b | 3.52 | $78,555 | 32.5 bps | 0.170 |
| xyz:XYZ100 | $7.11b | 0.67 | $564,449 | 4.9 bps | 0.045 |
| xyz:SKHX | $7.10b | 0.65 | $63,757 | 32.7 bps | 0.066 |
| xyz:SP500 | $6.56b | 0.82 | $700,843 | 4.5 bps | 0.072 |
| SOL | $6.43b | 3.98 | $151,125 | 21.2 bps | 0.109 |

Round trip is `2 × (taker fee + $1m ÷ depth)`, in basis points: what it costs to buy and sell the same million dollars of notional at the measured impact, fee included. R² is the fit of the lambda regression, and it is low on the thin rows for the obvious reason. Depth is measured per coin, so the rows above rank markets, not venues.

The result that should stop a comparison table is the HIP-3 row. A $1m round trip in the `xyz:SP500` perp costs 4.5 bps, and in `xyz:XYZ100` 4.9 bps, both cheaper than bitcoin at 8.8. The cheapest way to trade a million dollars per side in this field is a growth-mode equity index perp on Hyperliquid's builder book, not a major on any venue.

Over the 31 markets that trade more than $1b a month, the median round trip is 27.6 bps, the cheapest 4.5 (xyz:SP500) and the dearest 32.7. The spread inside one venue is wider than the spread between most venues' published fees, so a fee comparison measures almost nothing about cost.

## So which one

### Cheapest to trade?

Unanswerable from fee schedules, and that is the point of this post. 3 venues publish a zero taker fee and one lets a builder set your fee to zero, so a table sorted by its fee ranks venues that earn somewhere else. Hyperliquid is the only venue where the fee actually paid is measurable from public data, and there it is 2.86 bps, not the 4.5 bps on the fee page.

### Most listings?

Aster, with 641 perps and every one of them trading. Hyperliquid lists 565, of which 235 are core perps and 330 sit on ten builder-run venues, and 350 traded in the last 24 hours. Extended lists 416 and trades 242. If you want a market to exist, Aster; if you want the market to have size, neither is Hyperliquid.

### Deepest books?

Hyperliquid at the venue level: $1,456,452 of takedown moves bitcoin one basis point, and $741,139 does it for ether. But depth is a property of two markets, not a venue. The median coin above $1b a month takes $78,555, and a $1m round trip costs 4.5 bps in the HIP-3 S&P 500 perp against 32.5 bps in ZEC. Deep is a per-market answer everywhere in this field, including here.

### Most like Hyperliquid?

Lighter, at $1.84b and 202 trading markets, is the closest thing to a second venue of the same shape: the same 1h funding, an Ethereum L2, and 245 listed markets against Hyperliquid's 565. Its fee model is where the resemblance stops, since Hyperliquid charges a taker fee and Lighter charges standard accounts nothing.

### A small account?

Published schedules favour it almost everywhere: Hyperliquid's taker fee falls from 4.5 to 2.4 bps with volume and staking, while Paradex retail pays 0/0 and Lighter standard accounts pay nothing. What no schedule prices is the spread you cross on the way in, which is the number the zero-fee venues earn. Nothing here measures it.

### A venue that will still exist next year?

Weight the two that did not survive: Vertex shut down in July 2025 and Zeta stopped trading on 1 May 2025, and both still appear in comparison tables written before that. Of the survivors, 15 report volume at all and only a handful clear $100m a day. Size is not a guarantee, but a venue with no volume has nothing to lose by closing.

Two of the three questions have measurable answers; the third is a refusal grounded in how the field prices itself. That is the honest state of the field: listings and depth can be counted, and the fee that determines profit cannot be compared across venues because four of them earn it somewhere other than the fee line. Any table that ranks them by published taker rate is ranking their marketing.

How this table was checked

Market counts and 24h notional were probed from each venue's own public endpoint on 17 November 2026 at 13:16 UTC; the numbers are as of that minute and move constantly. CoinGecko's derivatives feed was read at the same time as a second opinion, and it agrees within 3.6% at worst (Extended, aggregator higher). Where a venue exposes no notional at all, the aggregator is the only record and the row says so. That is Aevo, Orderly, Nado, GMX (Arbitrum), Variational (Omni), edgeX, ApeX Omni, RISEx. Published fee schedules are read from each venue's documentation and are tier-0 rates rather than what any particular trader pays. edgeX's help centre and its VIP dashboard disagree on the perps fee as of this check (1.8/3.8 against 4.0/4.5 bps), so the row reproduces the help centre and flags the conflict rather than picking a winner. Aevo's and Nado's own endpoints carry no notional and Orderly's carries none either, so those rows lean on the aggregator. No one pays for a place in this table. Check the venue before you trade; fees and funding intervals move, and a roster that is a year old is worse than no roster.

[← Back to blog](https://tesseralytics.dev/blog)
