# Lighter vs Paradex: Two Zero Schedules, Two Different Books

Updated: 24 September 2026 — Tested: 22 September 2026 — Session Log staff

Lighter vs Paradex comes down to the book, because the schedule is the same: nothing to pay on the
tier most accounts trade, so a $10,000 round trip costs $0.00 either way. Neither
is on this log's register. Our medians read $16.2 million of resting depth against
$245.00. Leverage closes either position on a small move.

## Weights and scoring range

Cost of a round trip 30 · Measured liquidity 30 · Custody and settlement 15 · Security record 15 · Market coverage 5 · Access 5. Scored out of 25, one formula for the whole site, fixed on 22 September 2026: https://coinlen.com/how-we-rate

## Lighter vs Paradex, the lines that decide the bill

| Line | Lighter | Paradex |
|---|---|---|
| Score of 25 | 20.8 | 15.8 |
| What it is | Sequenced and matched off-chain, every batch proved to Ethereum | Matched in an off-chain cloud, settled on a Starknet appchain |
| Maker and taker on the tier most accounts trade on | 0% maker / 0% taker on a Standard account | 0% maker / 0% taker on anything sent from the web interface |
| What the zero tier asks in return | 300 ms added to a taker order and 200 ms to a maker order | A 300 ms speed bump, three orders a second and 1,000 a day |
| What a $10,000 taker round trip costs | $0.00 | $0.00 |
| Impact on orders of $10,000, $100,000 and $1 million | 0.43 bps, 0.71 bps, 1.55 bps | 113.92 bps, 160.58 bps, not filled |
| Documented BTC cap, and what bounds it | 50x, 2% margin | 50x, 300 BTC limit |
| Perpetual contracts listed | 214 | 63 |
| Getting out without the operator | An exit queue on Ethereum, with Desert Mode behind it | Nothing documented; a lockdown page lets the operator limit withdrawals |
| Open interest and 24-hour volume, as published by CoinMarketCap | $585.4 million and $2.0 billion on 22 September 2026 | $23.7 million and $21.8 million on 22 September 2026 |

Venue pages read 18 September 2026 in each venue's own documentation and terms.

## Both books while we watched

Polled six times an hour until the log closed on 22 September 2026; medians of all successful snapshots.

| Venue | $10,000 round trip | Spread | Depth, 10 bps | Impact, $100,000 | Snapshots |
|---|---|---|---|---|---|
| Lighter | $0.00 | 0.01 bps | $16.2 million | 0.71 bps | 52 |
| Paradex | $0.00 | 11.85 bps | $245.00 | 160.58 bps | 126 |

## Quoted, word for word

> "For both perpetual futures and spot markets, Lighter currently charges no maker or taker fees for Standard Accounts, allowing all participants to trade across all markets free of charge." — Lighter documentation, trading fees, 18 September 2026. https://docs.lighter.xyz/trading/trading-fees

> "Retail Takers and Makers have 0% fees on all products: Perps, Spot and Options" — Paradex documentation, trading fees, 18 September 2026. https://docs.paradex.trade/trading/trading-fees

> "Orders are only stored on the Paradex Cloud and are only known to the operators of Paradex Cloud." — Paradex documentation, privacy, 18 September 2026. https://docs.paradex.trade/trading/privacy

## [Lighter](https://lighter.xyz) — the book behind the zero

20.8 of 25. Points before the weights, each out of ten: Cost of a round trip 10.0 · Measured liquidity 6.4 · Custody and settlement 10.0 · Security record 7.0 · Market coverage 9.2 · Access 8.0.

Lighter sequences and matches orders off-chain and proves each batch of exchange operations to
Ethereum, which holds the deposit contracts and the state root
([Lighter architecture documentation](https://docs.lighter.xyz/about-lighter/technical-architecture-lighter-core.md), checked 18 September 2026). Its default
account pays 0% maker / 0% taker and waits 300 ms on a taker order for the privilege. The
BTC book held a median $16.2 million within ten basis points across
52 readings at a spread of 0.01 bps, and a $1 million
order moved 1.55 bps against it. Its 214 perpetual
contracts run from crypto to shares, currencies, commodities, indices, pre-IPO names and a
government bond.

- Works: Nothing to pay on the default account, across all 214 of its perpetual contracts; A documented exit queue on Ethereum, with Desert Mode behind it if the sequencer stops answering.
- Falls short: One operator-run sequencer on a cloud provider orders the entire book; The free account is deliberately the slow one, 300 ms on a taker order against 140 ms on the paid tier; No public bug bounty; reports go to a mailbox and any payout is at the venue's discretion.
- Not for: traders who need the quickest route to the book and would rather pay for it.

## [Paradex](https://www.paradex.trade) — the terms, rather than the book

15.8 of 25. Points before the weights, each out of ten: Cost of a round trip 10.0 · Measured liquidity 0.0 · Custody and settlement 7.0 · Security record 10.0 · Market coverage 7.0 · Access 8.0.

Paradex keeps its order book in an off-chain cloud that only its operators can read and
settles trades on Paradex Chain, a Starknet-stack network that posts proven state differences
to Ethereum ([Paradex privacy documentation](https://docs.paradex.trade/trading/privacy.md), checked 18 September 2026). Anything
sent from the web interface is a retail order at 0% maker / 0% taker, bought with a 300 ms
speed bump and no more than three orders a second. The book we read was the thin side of this
page: a median $245.00 within ten basis points across
126 readings, a spread of 11.85 bps and
160.58 bps on a $100,000 order.

- Works: A Sherlock bounty paying up to 500,000 USDC, behind four audit reports and a 2025 penetration test; 63 perpetual contracts, plus dated option contracts and spot in the same margin account.
- Falls short: The thinnest book we read on this desk, at 160.58 bps on a $100,000 order; A $1 million order ran past every published level on all 126 readings; If the fund behind liquidations runs dry, every withdrawal is cut by a socialised-loss charge; USDC is the only collateral it accepts.
- Not for: traders working six-figure clips, who pay in the fill what the schedule does not charge.

## Notes taken during the session

1. The fee argument ends at once: $0.00 for a $10,000 taker round trip on either side, and the same after twenty of them.
2. The fill argument does not. A $10,000 order moved 0.43 bps against one book and 113.92 bps against the other: $0.43 of slippage against $113.92.
3. At $100,000 that is $7.10 against $1,605.80 on one leg, so a schedule of nothing costs more here than any schedule on this desk.
4. A $1 million order filled at 1.55 bps on one side; on the other it ran past every level published, on all 126 readings.
5. Neither column is on the register this log ranks: it admits an entry tier of five basis points or less to cross, and a zero bought with 300 ms of delay is not that.

## What a schedule of nothing asks for in return

Neither venue charges the account an ordinary trader opens. Lighter states that Standard
accounts pay no maker or taker fee across every market it lists
([Lighter fee documentation](https://docs.lighter.xyz/trading/trading-fees.md), checked 18 September 2026), and Paradex has priced
retail orders at nothing on perpetuals since 15 June 2026
([Paradex fee documentation](https://docs.paradex.trade/trading/trading-fees.md), checked 18 September 2026). Both then charge for
the thing a fee usually buys, which is speed.

On Lighter the free account waits 300 ms on a taker order and 200 ms on a maker order, while
the opt-in Premium account waits 140 ms and pays 0.004% and 0.028% for the difference. On
Paradex an order counts as retail by how it was sent, from the web interface or an interactive
token, and retail orders meet a 300 ms speed bump, with three orders a second and a thousand a
day as the ceiling. Sent the standard programmatic way instead, the same order costs 0.003% and
0.045%.

Both venues price that delay themselves, which says what it is worth: Lighter sells 140 ms for
0.028% a crossing, Paradex lifts the speed bump for orders paying 0.045%. So the zero is not quite
a zero, and neither column is on the register this log keeps, which admits an entry tier of five
basis points or less to cross. A free trade that arrives late is not free to a trader working the
same day. The rest of this page is about the other half of the bill.

## The bill that arrives with the fill

A market order pays the book, not the fee page, and the two books here are not comparable.
Across snapshots taken every 10 minutes until the log closed on
22 September 2026 — 52 of one book and
126 of the other — one BTC book held a median $16.2 million
within ten basis points of the mid and the other $245.00 — a factor of about
66,059, on the same asset, at the same minute.

That difference is the whole price. A $10,000 market order moved
0.43 bps above the mid on the deeper book and
113.92 bps on the thinner one: $0.43 of slippage against $113.92, or
$0.86 against $227.84 once the position is closed. At $100,000 it stops being loose
change, at 0.71 bps against 160.58 bps, or $7.10
against $1,605.80 on one leg.

At $1 million the thinner book stops answering. The deeper one filled the whole order at
1.55 bps, about $155.00 of slippage on one leg; on the other the order
walked through every level published and was still unfilled, on all
126 readings.

## What each one does when the sequencer stops

Both books are ordered by one operator, so what matters to collateral left overnight is what a
trader can do when that operator goes quiet. Lighter documents an answer: exits go straight to
Ethereum through a priority queue, and if the sequencer ignores them the contract freezes and
traders leave with proofs
([Lighter architecture documentation](https://docs.lighter.xyz/about-lighter/technical-architecture-lighter-core.md), checked 18 September 2026). Nethermind audited
that path and the deposit bridge in September 2025, two of the nine reports it publishes.

Paradex documents the opposite arrangement. Its book is held in a cloud its operators can read
([Paradex privacy documentation](https://docs.paradex.trade/trading/privacy.md), checked 18 September 2026), no forced withdrawal
appears in the pages read on 18 September 2026, a lockdown page lets the operator restrict
them, and a native-bridge withdrawal waits about two hours for a Starknet proof.

Each record carries one heavy entry. Lighter was down four and a half hours during the crash of
10 October 2025, blamed its own sequencer database and compensated traders; Paradex published
corrupted state on 19 January 2026 and refunded $650,000 to 200 accounts. Money came back both
times, so the bounty line is what separates them.

## FAQ

### Lighter or Paradex for day trading?

On this desk Lighter takes 20.8 of 25 against
15.8, and almost all of the gap is the order book: a median
$16.2 million resting within ten basis points against $245.00. Paradex
scores higher on the security line, with a published bounty. For an intraday session the book
decides.

### Do Lighter and Paradex really charge no fees?

On the tier most accounts trade on, yes. Lighter charges Standard accounts nothing on every
market, and Paradex has charged retail orders nothing on perpetuals since 15 June 2026. Both
run paid tiers alongside: 0.004% and 0.028% on one, 0.003% and 0.045% on the other, bought for
lower latency and higher limits.

### What did a $100,000 market order cost on each?

Against the books as we read them, 0.71 bps on the deeper one and
160.58 bps on the thinner one, which is $7.10 of slippage against
$1,605.80 on a single leg. Enter and leave and that is $14.20 against $3,211.60, with neither
venue charging a fee.

### Why is one order book so much thinner than the other?

Size follows size. On 22 September 2026 CoinMarketCap listed $585.4 million of open interest
and $2.0 billion of 24-hour volume on one venue, against $23.7 million and
$21.8 million on the other. Market makers rest capital where flow arrives, and our
snapshots read the result rather than the cause.

### What is Desert Mode on Lighter?

The emergency exit under its contracts. Traders can put withdrawals and reduce-only exits into
a priority queue on Ethereum; if the sequencer does not process them inside the published
window the contract freezes and balances leave with proofs. Nethermind audited that path in
September 2025, and no equivalent appears in the Paradex pages we read.

### What happened to Paradex on 19 January 2026?

A race condition during planned database maintenance published corrupted state to its chain,
the funding index reset to zero and accounts were liquidated that should not have been.
Paradex rolled the chain back to an earlier block, the first rollback in its history,
cancelled open orders and refunded $650,000 to 200 accounts.

### Can I get my money out if the operator stops answering?

On one of the two, by documented design: exits join a priority queue on Ethereum, and the
contract freezes if it ignores them. On the other there is no forced withdrawal in the
documentation read on 18 September 2026, native-bridge withdrawals wait about two hours for
a proof, and a lockdown page lets the operator restrict them.

### How many markets does each one list?

Counted from their own public endpoints on 18 September 2026,
214 active perpetual contracts on one and 63 on the
other. Neither number decides a session, because intraday volume sits in a handful of pairs.
EVEDEX lists 52, fewer than either, and its documentation, read on 18 September 2026,
describes no way out past the operator.

### Does either one ask for identity documents?

Neither, in the onboarding pages read on 18 September 2026. Lighter registers a main account
from a signed message and takes a deposit through its Ethereum contract; Paradex derives an
account from the wallet address itself. Both sets of terms still exclude whole countries.

### Is a fee schedule of nothing worth a slower order?

Both venues answer that by selling the faster lane: 140 ms for 0.028% a crossing on one, and no
speed bump at 0.045% on the other. On this log the answer is no, and it is why neither sits on
the register: a session of forty crossings pays the delay forty times, in the move rather than in
the schedule.

## How the session was run

Two venues, the six criteria this desk fixed on 22 September 2026 and uses on every page here
([how we rate](/how-we-rate)). Fee schedules, market counts, margin fractions, restricted
countries, audit records and incident post-mortems come from each venue's own documentation,
terms and public endpoints, read on 18 September 2026. Spread, resting depth and price impact
are ours: both BTC perpetual books polled six times an hour until the log closed on 22 September 2026,
52 readings of one and 126 of the other, reported as
medians. Publication here is paid for.

Neither venue is ranked anywhere else on this site, so nothing here places them against the seven on
the register. One session on BTC alone is a thin sample: a book that holds in a quiet hour can
empty in a fast one. Price impact is read against the levels each venue publishes, so one that
shows fewer looks worse. A schedule of nothing still ignores funding, charged for every hour a
position stays open. The wider field is on the [venue list](/crypto-day-trading-platform-list).

This site is paid placement. Both columns are scored with the published formula and nothing else, and every figure carries its date and its source. Corrections: editorial@coinlen.com.

Session Log staff, 24 September 2026

This site is paid placement.