Tools

What a leveraged trade actually costs you

Eleven of the twelve products below charge a share of your notional, twice, whether you win or lose. Papertrade charges a share of your gain, once, and nothing if there is no gain. No single fee rate can hold both, so this computes what each one takes from the same trade. Drag the price move: the order changes, and where it changes is the answer.

Cohort-wide rankings of taker fee, spread and impact live on the perp DEX leaderboard, and the maximum-leverage terms behind the liquidation column are the leverage terms benchmark.

Market
Jump to
0.200% of BTC price= +100% on your margin, $1,000 on $500,000 of notional
Papertrade logo

Cheapest on this trade

Papertrade

charges on the gain · HyperEVM

It takes
$195
19.5% of the gain
You keep
$805
80% on your margin
Versus next
$4.99
cheaper than Gains x500 degen, fixed
Open here
3 of 12
products at 500x

Why the order flips

A flat fee is a line, a rate that climbs is a curve

$0$200$4000%0.10%0.20%0.30%0.40%0.50%0.60%Gains x500 degen, fixed $200crosses at 0.208%you are here
A fee on notional does not move with the price, so it draws a flat line. The curve is not simply a share of the gain, which would be a straight line too: it bends because the share itself climbs with the move, from nothing at the deadband toward 11.8%. They cross once. Framed on the region where the two are comparable, up to a 0.62% move; the slider runs to 2%.
  • Papertrade logoPapertradecharges on the gain$195
  • Gains logoGainsx500 degen, fixed$200
  • Nova logoNovax500 degen, fixed, same terms as Gains$200

The winning path

What each venue keeps from your gain

Cost, net gain, share of the gain and room to liquidation, by venue
RankVenueCostYou netShare of gainRoom to liq.
1
Papertrade logo

Papertradecharges on the gain

HyperEVMHouse LP

$195$80519.5%0.152%of price · api
2
Gains logo

Gainsx500 degen, fixed

ArbitrumOracle pool

$200$80020.0%0.120%of price · chain
3
Nova logo

Novax500 degen, fixed, same terms as Gains

ArbitrumOracle pool

$200$80020.0%0.120%of price · chain · 2026-10-22
n/a
Gains logo

Gainsstandard pair

ArbitrumOracle pool

not offered at 500xcaps at 200x
n/a
Nova logo

Novastandard pair

ArbitrumOracle pool

not offered at 500xcaps at 200x
n/a
Lighter logo

Lighter

zkSync rollupOrder book

not offered at 500xcaps at 50x
n/a
Paradex logo

Paradex

StarknetOrder book

not offered at 500xcaps at 50x
n/a
Extended logo

Extended

StarknetOrder book

not offered at 500xcaps at 50x
n/a
edgeX logo

edgeX

StarkExOrder book

not offered at 500xcaps at 100x
n/a
Hyperliquid logo

Hyperliquid

Hyperliquid L1Order book

not offered at 500xcaps at 40x
n/a
dYdX v4 logo

dYdX v4

dYdX ChainOrder book

not offered at 500xcaps at 50x
n/a
GMX v2 logo

GMX v2implied cap

ArbitrumOracle pool

not offered at 500xno published cap; the margin requirement bites below 200x

Against Gains x500 degen, fixed, the cheapest notional-fee venue open at 500x, the crossover sits at a 0.208% move, a gain of 104% on your margin. Below it Papertrade takes less; above it Gains does, and the gap widens from there. Against this one rival the leverage does not move the crossover, because it multiplies both costs equally. What leverage changes is which rival is open at all, and that moves the number a long way.

The other two paths

What happens when it does not work

What a flat close and a liquidation cost, by venue
VenueTrade closes flatTrade is liquidated
Papertrade logo

Papertradecharges on the gain

costs nothingnothing comes backmints PAPER, about 29 per $1 of loss basis
Gains logo

Gainsx500 degen, fixed

$200 of fees on a $0 gainnothing comes back
Nova logo

Novax500 degen, fixed, same terms as Gains

$200 of fees on a $0 gainnothing comes back

Only the venues open at the chosen leverage appear here; the first table lists the others and why. The liquidation column is the equity the venue's own trigger leaves behind, before the closing fee that trigger charges, and it is not a promise the money reaches you. On Gains it does not: we paired every liquidation we could find and amountSentToTrader read zero on all 27, across every pair from 7.5x to 500x. Papertrade's documentation calls its liquidation a hard bust that forfeits the full isolated margin. For the others this is the parameter, which we have not checked against a realised payout.

One thing does come back on Papertrade, and it is not margin. Losing is the only event that mints its token: the docs state that PAPER mints to your account whenever you close a trade at a loss or get liquidated, and nothing else mints PAPER. The rate is published in full. Below $2M of tracked LP it is a flat 100 PAPER per dollar of mint basis; at or above it the rate is 100 × (S ÷ (S + H))², where S is the $120M decay scale and H a high-water-mark of cumulative LP gain past the threshold that never falls back. H is not published as a field, so it is read off the peak of the venue's own daily tracked-LP series, $104.89M, which puts the rate near 29 PAPER per dollar of loss; the venue's own worked example, 29.8 at a $100M high-water-mark, lands beside it. Because H only ratchets upward, that rate falls for later traders and never recovers. A liquidation mints on the full lost margin; an ordinary losing close mints on the loss after the 2% fee, which is carved from the LP's side rather than added to the loss. Two things stop any of it being a rebate: at launch PAPER cannot be transferred, only staked and unstaked, and what staking pays is a pro-rata share of future protocol revenue that dilutes as later losers mint more. The row says both that no margin returns and that something is minted, because either sentence alone is misleading.

How these numbers were read

Fees and spreads come from our own perp-fees measurements, read on 2026-10-21, except the few rows that carry a later date of their own beside their provenance tag. Each spread is the median of 10,080 readings, one a minute over seven days, and half of it is charged per leg because a taker crosses half the book. Readings at or below zero are dropped first: a crossed or empty book is the reader failing rather than a tighter market, and Paradex produced 55 of them in that window. Until 21 October each spread was instead a single tick, which is a sample of size one from a quantity that moves several-fold inside a minute, and it decided the whole figure on the two rows whose taker fee is zero. It also went wrong quietly: dYdX carried 8.4248 bps on BTC against a seven-day median of 0.8495, a value that never occurred on the day the page said it was read. Paradex also publishes two taker tiers on the same market: its fee_config reads zero for an interactive order and 0.0200% a leg for an API or RPI one. The interactive tier is shown, because this page prices one trade for a person rather than for a programme. Two rows carry no spread, for different reasons: Papertrade quotes one price and charges nothing to cross it, and GMX v2 has no book at all, so what a taker pays there is price impact against the pool, which this page excludes. Maintenance requirements are marked by provenance in the liquidation column: api read from the venue's own endpoint, chain read from its contracts, docs derived from a documented rule rather than a published field. The figures are fixed in the page at that date, not polled live.

Gains, Arbitrum diamond 0xFF162c69…F169: pairs() word 9 gives the fee index and fees() word 0 the position-size fee at 1e10. Standard pairs read a 0.0350% taker a leg with a 0.0100% pair spread and run 1.1x to 200x; the x500 degen pairs read 0.0200% with no spread, and both pairMinLeverage and pairMaxLeverage read 500x, so that product is fixed at 500x rather than capped there. The liquidation threshold also comes off chain, from getPairLiquidationParams(): standard pairs give up 90% of collateral at or below 25x and 75% at or above 60x, interpolated linearly in between, while the degen pairs read 80% flat. The fee sits inside that trigger, so the room shown subtracts both fee legs from it. On the degen pairs the 80% threshold models a room of 0.1200% at 500x, which is the figure the table renders; eleven real liquidations fired at a median 59.4% of collateral instead, which works out at 0.1189%.

Nova, Arbitrum diamond 0x1410FA3F…34f3: the same five reads on a deployment of the same gTrade contract, so the thresholds, the leverage bounds and the pair indices are identical to Gains and only the pricing differs. Its standard pair reads a 0.0500% taker a leg and no spread at all, against Gains' 0.0350% with a 0.0100% spread, which is 5.0 bps a leg against 4.0 once the half-spread is counted on both sides. Word 7 of pairs() is the pair spread at 1e10 and reads zero on Nova, so the row carries zero rather than Gains' value: copying the neighbouring figure would have put Nova at 5.5 bps and looked right beside the identical thresholds. The x500 degen pairs are the same product on both, 0.0200% a leg at a flat 80% threshold, which is why that row says so. Read on 2026-10-22, a day after the rest of this page, and both Nova rows carry that date beside their provenance tag.

Papertrade's haircut is its own published formula, scale = (1 - baseRate) / (1 + 1/(move x rateMultiplier) + referenceNotional/(10^6 x move x positionMultiplier)), with baseRate 10%, rateMultiplier 15,000, referenceNotional $100,000 and positionMultiplier 814.598 on BTC and 483.979 on ETH, then a 0.2 bps deadband and a 2% win fee. Its bust buffer comes from /state/trading and is an offset before the zero-equity price, so the room is one over the leverage minus the buffer, not the buffer itself. The haircut uses a constant reference notional rather than your position size, by design, so it does not vary with how much you trade.

Not shown. Aster: its leverage-bracket endpoint needs a key, and the only public margin fields contradict the product it advertises. Ostium: its subgraph publishes two threshold fields and only one matches its documented formula, an unresolved conflict. A missing row beats a wrong one on a page of money figures.

Excluded from every figure: funding, borrowing, GMX's price impact, and the closing fee and spread charged at liquidation. All of them move the real trigger closer than shown, and the longer a position is held the more they matter. edgeX and GMX also tier their maintenance requirement by position size and open interest; the base tier is used here. The model covers 12 products across 10 venues, on BTC and ETH only.