# How we rate

Updated: 24 September 2026. The Cost Per Trade pricing desk

## In short

Every sheet on this site is priced with one methodology, fixed on 22 September 2026
before any venue was scored. A page condition — an order size, a fee model, a brand being replaced —
decides which venues appear on a sheet. The weights below decide the score, and
[what orders each sheet](#what-orders-each-sheet) says which number sets the sequence. Scores run from
zero to 50 and are rounded to 1 decimal place.

## Criteria and weights

| Criterion | Weight | What it measures | How it becomes points |
|---|---|---|---|
| Published rate | 40 | What the rate card alone charges for one round trip in the BTC perpetual — the entry-tier taker rate, paid going in and paid again coming out, in basis points. | A round trip billed at two basis points, or at nothing more than two, keeps the full ten points. Every basis point above that costs the line half a point, so the ten is spent by twenty-two — charge nothing on either side and all ten stand, charge 0.045% a side and nine basis points in and out leave six and a half. |
| Execution cost | 30 | The one price on this sheet no venue quotes: what stands in the BTC book within 10 basis points of the mid, buy side and sell side at once, read by this desk and reduced to a median. | Depth is priced the way a wholesale sheet prices quantity, in breaks of ten. The first break is $10,000 standing in the band, and it buys nothing; each break of ten after it pays two points more, so $100,000 carries two points, $1 million four of them and $10 million six. The column stops paying at ten. |
| Where the money sits | 12 | What the money is doing while the position is open — which parts of a trade are posted to a chain, who may move the deposit, and who has audited the contract it sits in. Read off the venue's own pages, not tested here. | the fills themselves reach a chain, in rollup batches or under a validity proof — 4 points; the fills stay off the chain and only the open position is posted, at intervals — 2 points; the paperwork sets out how to get the collateral back without asking the operator — 3 points; an audit has been published on the contract the deposit lands in — 2 points; nothing is opened, since a wallet signature does the signing in — 1 points |
| Security record | 8 | The record standing behind the rate: audits published by somebody outside the venue, a bug bounty open to anybody, and what the past 24 months hold against it. | an audit from outside the venue on the record, and a second one worth as much again — 4 points; a bug bounty that pays outsiders for what they find — 3 points; no trader money gone in the past 24 months, by the public record — 3 points |
| What the rate buys | 6 | What the price buys access to: the number of perpetual contracts on a venue's own documentation, and how many kinds of asset beyond crypto sit on that list. | The contract count is worth six points, priced on a curve rather than a line: five hundred perpetual markets pays the six in full, and most of the six has been handed over well before a list gets that long. The last four points are sold one to an asset class outside crypto — commodities, currencies, indices, shares, pre-IPO names — and no more than four are counted. |
| Who may pay it | 4 | Who is allowed to pay the rate at all: the identity papers demanded before a first order, what it takes to sign in, and how long the refused list runs. Five points, doubled to sit on the same ten-point scale as the rest. | no papers about who you are, at any point before a first order — 2 points; a signature from a wallet is the whole of the sign-up — 1 points; the terms turn away sanctioned parties and one large market, and no one else — 2 points; the refusal list runs longer than that — 1 points |


Every criterion is scored out of ten first. The weights are then applied, the total is rescaled to
run from zero to 50, and the result is rounded to one decimal place. The points
for who may pay it are counted out of five and doubled, so that criterion also lands on a ten-point
scale before it is weighted.

## A worked example

Take EVEDEX on 18 September 2026. Its published taker rate is 0.045%, so a round
trip costs nine basis points; the published-rate criterion starts at ten points for two basis points
or less and removes half a point for each further basis point, which leaves six and a half. Its BTC
book held $18.3 million within 10 basis points, above ten million, so the execution-cost
criterion clears the six points that ten million pays and lands on 6.5.

Position data reaches a chain periodically rather than trade by trade, there is no documented exit
without the operator, the deposit contract is audited and sign-in is by wallet: two plus nothing plus
two plus one is five points for where the money sits. Two audits and no loss of user money in
24 months, with no bug bounty, give seven for the security record. Fifty-two markets and five asset
classes beyond crypto give 7.8, and no identity documents plus a wallet signature plus restrictions
wider than sanctions give four of five, doubled to eight.

Multiply each by its weight, add them up, divide by the weights and rescale: 32.5
of 50.

## What every venue scored

Every score here stands as of 18 September 2026 and rebuilds from the columns beside it. A round
trip is the entry-tier taker fee charged twice, once on the way in and once on the way out; depth is
the median of what our BTC readings found waiting within 10 basis points of the mid price.

| Venue | Score | Published rate | Execution cost | Where the money sits | Security record | What the rate buys | Who may pay it | Round trip | Depth | Settlement | Exit without the operator | Audits | Bug bounty | Loss in 24 months |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Lighter | 42.8 | 10.0 | 6.4 | 10.0 | 7.0 | 9.2 | 8.0 | 0.0 bps | $16.2 million | every trade | yes | 9 | no | no |
| Extended | 38.4 | 8.5 | 5.8 | 7.0 | 10.0 | 9.6 | 8.0 | 5.0 bps | $7.8 million | every trade | no | 2 | yes | no |
| Aster | 36.3 | 7.0 | 6.3 | 7.0 | 10.0 | 10.0 | 8.0 | 8.0 bps | $14.9 million | every trade | no | 7 | yes | no |
| edgeX | 36.2 | 6.5 | 6.9 | 10.0 | 7.0 | 8.0 | 8.0 | 9.0 bps | $29.2 million | every trade | yes | 6 | no | no |
| Hyperliquid | 33.1 | 6.5 | 5.7 | 7.0 | 7.0 | 9.6 | 8.0 | 9.0 bps | $7.3 million | every trade | no | 2 | yes | yes |
| EVEDEX | 32.5 | 6.5 | 6.5 | 5.0 | 7.0 | 7.8 | 8.0 | 9.0 bps | $18.3 million | periodic | no | 2 | no | no |
| Paradex | 31.9 | 10.0 | 0.0 | 7.0 | 10.0 | 7.0 | 8.0 | 0.0 bps | $870.00 | every trade | no | 5 | yes | no |
| ApeX Omni | 31.8 | 6.0 | 4.4 | 10.0 | 7.0 | 7.7 | 10.0 | 10.0 bps | $1.7 million | every trade | yes | 4 | no | no |
| dYdX | 28.0 | 6.0 | 3.9 | 7.0 | 7.0 | 5.2 | 8.0 | 10.0 bps | $853,419 | every trade | no | 6 | yes | yes |
| Aevo | 22.4 | 3.0 | 4.3 | 7.0 | 4.0 | 8.5 | 8.0 | 16.0 bps | $1.5 million | every trade | no | 5 | no | yes |

## What counts as a loss of user money

Money counts as lost when the people who put it there did not get it back: drained in an exploit, drained out of a vault the venue itself ran by way of price manipulation, or stranded by an outage and never made whole afterwards. A position liquidated because the market went the other way costs its trader money, and is not a loss in this sense.

## Which venues we price

Scoring covers venues that match perpetual futures on a central limit order book, whichever side of the chain that book happens to live on, and whose published entry-tier taker rate is 0.045% or less: seven venues clear that line. Three order-book venues charge more than it — Aevo at 0.08%, ApeX Omni and dYdX at 0.05% — and are quoted where they set the top of a range without being scored. Where the price comes from a pool or an oracle instead, the venue keeps its published charges on the fee list and takes no score, there being no book for the execution-cost criterion to read.


Venues that are listed but not priced keep their published charges on
[the crypto exchange fee list](/crypto-exchange-fee-list), with a note saying why no score is given.

## Where the numbers come from

- **The printed half of the bill** — rates, market counts, sign-in rules, restrictions, audits — is
  read off each venue's own rate pages, documentation and terms of use, on the day shown beside the
  figure. The entry tier is the charge a fresh retail account meets on the venue's own interface
  with nothing yet working in its favour: no turnover behind it, nothing staked, nobody referred and
  no cashback.
- **The unprinted half is timed here.** A script goes to the public BTC perpetual order book of
  every venue priced on this site, once each 10 minutes. Depth adds up the orders waiting within
  10 basis points of the mid price on both sides; the walk measures how far the average fill on a
  market order of a stated size lands from that mid price. What the tables carry are medians, and
  the reading count sits with them.
- **Anything from outside** is taken at the source and nowhere in between: open interest and
  24-hour volume as CoinMarketCap publishes them, audit records as CertiK Skynet holds them. No line
  on these sheets is copied off a rival comparison table.
- **The all-in price** on every sheet is the published fee on a $10,000 round trip added to what the
  order-book readings show the book taking on the way in and again on the way out.

## What orders each sheet

Which venues reach a sheet is settled by what the page is about, and so is the order they stand in.
Where a sheet is built on a single measurement, that measurement sets the sequence; everywhere else
the sequence comes from the overall score.

| Sheet | Ordered by |
|---|---|
| Home, Lighter alternatives, Low fees, scored, Fee models | All six criteria priced in and weighted, biggest total first |
| No maker fee, Cheapest first, The whole tariff | What the rate card alone takes out of a $10,000 round trip, least first |
| Large orders | What was standing in the BTC book, within 10 basis points of the mid, when it was read |


Where a measurement sets the order, the overall score keeps its column all the same, and two venues
level on the measurement are separated by it. Criteria, weights and the scores they produce hold
still across the site: whatever a venue scores on one sheet is what it scores on every other.

## Ties and rounding

Venues whose totals are equal after rounding share a line and are marked as sharing it. The sequence inside a shared line is alphabetical and means nothing.

## Who paid for this method

A method is worth reading only when the reader knows who financed the page it sits on. These sheets
were bought, and a venue scored by the formula above may be the buyer. The weights were settled
on 22 September 2026, before a venue had been priced against them,
and nothing in the arithmetic was reopened afterwards for anyone. The rest of the arrangement is on
[about](/about).

## Updates

The rate card is read again once a week. The writing around it is redone when a venue changes its
schedule rather than on any calendar of ours. A new version of this methodology takes a new number
and a new date on this page, and the old sheets are rebuilt on it rather than left mixed in with it.

24 September 2026: first published version, fixed before any venue on this site was priced.

The Cost Per Trade pricing desk, 24 September 2026. Corrections: desk@thundermessage.com.

This price desk is paid to publish, and a venue on the sheet may be the party that paid; the order of every column is the published formula.