# How we rate perp DEXs and hybrid exchanges

Updated: 24 September 2026 / Spaces Research

Every ranking on this site uses one methodology, fixed on 18 September 2026 before
any venue was scored. Page conditions — a chain, an audience, a leverage limit — decide which venues
appear on a page; the weights and formulas below decide the overall score, and [which metric orders each
page](#which-metric-orders-each-page) says which number sets the order. Scores run from 0 to
10 and are rounded to 1 decimal place.

## What the meter reads, and what each reading is worth

| Criterion | Weight | What the criterion measures | How it becomes a score |
|---|---|---|---|
| Trading cost | 25 | The cost column — a BTC perpetual opened at the taker rate a new account is quoted and unwound at the same rate, the two fills added and printed in basis points. | The meter starts at ten and holds there while a round trip clears in two basis points or less. After that it gives up half a point for every further basis point — six basis points reads eight, twelve reads five — and it bottoms out at zero once the round trip reaches twenty-two. |
| Liquidity | 25 | The one column on this screen we take ourselves — dollars of bids and asks resting within 10 basis points of the BTC mid, printed as the median of the run with its count of readings beside it. | Read in decades rather than in dollars. Ten thousand dollars resting inside the band registers nothing at all; each tenfold above it is worth two points, which puts a hundred thousand at two, a million at four, ten million at six. Ten, the top of the gauge, would take a billion dollars inside the band. |
| Custody and settlement | 20 | Nothing to poll here — the points are marked off the venue's own settlement, withdrawal and audit pages, ten of them at most. | each fill is written to a chain, inside a rollup batch or behind a validity proof: 4; a chain sees position balances now and then, and never the fills: 2; the documents describe an exit a trader can trigger without the venue: 3; a published audit covers the contracts the deposit sits in: 2; there is no exchange account to open, only a wallet that signs: 1 |
| Security record | 15 | No reading of ours goes into this one — the audit reports on the record, the bounty page, and what the last twenty-four months hold against the venue, out of ten. | a security audit on the record is worth two points, and only the first two count: 4; a bug bounty the venue keeps open to outsiders: 3; nothing gone from user balances in the twenty-four months behind the reading: 3 |
| Market coverage | 10 | What the venue puts within reach of one account — the perpetual contracts its own documentation and trading terms list, and the asset classes behind them outside crypto. | The contract count is logged rather than counted straight. Five hundred perpetual markets fill the six points on offer, fifty still stand near four, and nothing above five hundred adds more. On top of that sits a point for every asset class outside crypto — shares, currencies, commodities, indices, pre-IPO contracts — and the fourth is the last one counted. |
| Access | 5 | Five points read off the sign-in and access terms, then doubled onto the ten-point scale. | trading starts without an identity check: 2; a wallet signature is the whole sign-in: 1; a narrow exclusion list, no wider than sanctions and one further market: 2; an exclusion list wider than that: 1 |


Access points (0 to 5) are doubled to a 0–10 scale. Market coverage adds a logarithmic score for the
number of perpetual markets (500 markets reach the 6-point ceiling) to one point per asset class
beyond crypto — stocks, currencies, commodities, indices, pre-IPO contracts — up to 4 points.

## Current scores

Scores as of 18 September 2026, rebuilt from the inputs below with the formulas above. The first table holds
the total and the six criterion scores that make it; the second holds the raw readings those scores were
calculated from, so any row can be recomputed by hand.

### Total and criterion scores

| Venue | Score | Cost | Liquidity | Custody | Security | Coverage | Access |
|---|---|---|---|---|---|---|---|
| Lighter | 8.5 | 10.0 | 6.5 | 10.0 | 7.0 | 9.2 | 8.0 |
| Extended | 7.8 | 8.5 | 5.7 | 7.0 | 10.0 | 9.6 | 8.0 |
| Aster | 7.7 | 7.0 | 6.4 | 7.0 | 10.0 | 10.0 | 8.0 |
| edgeX | 7.6 | 6.5 | 6.9 | 10.0 | 7.0 | 8.0 | 8.0 |
| ApeX Omni | 6.9 | 6.0 | 4.5 | 10.0 | 7.0 | 7.7 | 10.0 |
| Hyperliquid | 6.9 | 6.5 | 6.0 | 7.0 | 7.0 | 9.6 | 8.0 |
| EVEDEX | 6.5 | 6.5 | 6.5 | 5.0 | 7.0 | 7.8 | 8.0 |
| dYdX | 5.8 | 6.0 | 3.9 | 7.0 | 7.0 | 5.2 | 8.0 |


### The readings behind them

Round trip is twice the entry-tier taker fee; depth is the median of the BTC order-book snapshots within
10 basis points.

| Venue | Round trip | Depth | Settlement | Exit without operator | Audits | Bug bounty | Loss in 24 months |
|---|---|---|---|---|---|---|---|
| Lighter | 0.0 bps | $16.9 million | every trade | yes | 9 | no | no |
| Extended | 5.0 bps | $7.4 million | every trade | no | 2 | yes | no |
| Aster | 8.0 bps | $16.1 million | every trade | no | 7 | yes | no |
| edgeX | 9.0 bps | $29.5 million | every trade | yes | 6 | no | no |
| ApeX Omni | 10.0 bps | $1.8 million | every trade | yes | 4 | no | no |
| Hyperliquid | 9.0 bps | $9.6 million | every trade | no | 2 | yes | yes |
| EVEDEX | 9.0 bps | $18.3 million | periodic | no | 2 | no | no |
| dYdX | 10.0 bps | $845,322 | every trade | no | 6 | yes | yes |

## When this desk records money as gone

User funds count as lost when an exploit or a hack takes them, when a vault the venue itself runs is emptied by market manipulation, or when an outage leaves them gone — in every case only where the venue has not made the holder whole again. A position liquidated by an ordinary market move is not a loss here.

## Which venues we score

Venues with a central limit order book (hybrid or fully on-chain) whose BTC book filled a $1 million market order in at least 95% of the readings of the window: liquidity carries a quarter of the score here, so a book that cannot fill that size is left off the table. Eight venues clear it — seven of them found a price for that size in every reading, and Hyperliquid in 112 of its 114. Aevo found one in 107 of its 113 readings, and the Paradex book found none at that size in any of its 114, so neither carries a total here, and nor do the oracle- and pool-priced venues, which have no resting book for the liquidity metric to measure at all.

## Data sources

- **Fees, markets, leverage, restrictions, audits:** each venue's documentation and terms of use,
  checked on the date shown next to the number. "Entry tier" means what a new retail account pays by
  default on the venue's own interface, before volume, staking or referral discounts.
- **Liquidity:** snapshots of the public BTC perpetual order book of 28 venues, taken every
  10 minutes. Depth counts resting orders within 10 basis points of the mid price on both sides; the
  page reports the median and the number of snapshots.
- **Third-party data:** CoinMarketCap (open interest and 24-hour volume), CertiK Skynet (audit
  records). Perpetual market counts come from each venue's own documentation and trading terms.

## Which metric orders each page

Page conditions decide which venues appear on a page. What sets their order depends on what the page is about:
a page built around one measurement is ordered by that measurement, and every other page is ordered by the
overall score.

| Page | Ordered by |
|---|---|
| Home, Best perp DEX for beginners, Perp DEX list, Best hybrid crypto exchanges, Aster alternatives | The weighted total, highest reading first |
| Best DEX for leverage trading | The documented BTC cap, and the notional it survives |
| Most liquid perp DEXs | Median dollars resting within 10 bps of the BTC mid |
| Perp DEX fees compared | What two taker fills cost on $10,000 at the entry rate |


On a page ordered by one measurement the overall score stays in the table, and equal values are broken by it.
The criteria, weights and formulas above never change between pages, and neither do the scores: a venue carries
the same score everywhere on the site.

## Totals that land on the same number

Totals that round to the same number share a place on the table and are flagged as tied. The order inside a tie is alphabetical and carries no meaning.

## Who financed the method

A formula means little until the reader knows who paid for the page carrying it. Publication on
this site is commissioned, and a venue scored by the formula above may be the sponsor behind it. The
weights were fixed on 18 September 2026, before any venue was scored against them, and no score has
been adjusted by hand since. The arrangement is described on [About Spaces Research](/about).

## Updates

A table is re-read when a venue changes its rate card, its leverage ladder or its market list, and
when a fresh run of order-book snapshots replaces the one a median was taken from. A new methodology
version gets a new number and date on this page; old rankings are rebuilt with it rather than mixed.

24 September 2026: the ordering rule for each page was written down in "Which metric orders each page", and the
leverage ranking moved from the overall score to the documented BTC leverage cap. Criteria, weights, formulas
and every venue score are unchanged from the 18 September 2026 version.

Spaces Research, 24 September 2026. Corrections: editorial@spacesdao.org.

Publication on this site is commissioned and paid for, and a venue compared here may be the sponsor behind the page.