Skip to content
PerpFinder

Compare perp DEX security (2026)

Short answer: This ranking sorts recorded editorial security scores. A score does not measure the probability of a loss or establish which venue is safest. Check each audit scope and the current controls.

Recorded security scores by venue

#ProtocolSecurity scoreFee modelMax levRating
1dYdX9.2/100.01% / 0.05%50x9/10
2GMX9/100.04% or 0.06% when opening and again when closing; rate depends on whether the action improves or worsens pool imbalance100x8.5/10
3Hyperliquid8.8/100.015% / 0.045%40x9.2/10
4Jupiter Perps8.5/100.06% when opening and again when closing; base position fee; price impact, borrowing, swap, and network costs are separate250x8.3/10
5Gains Network8.5/100.035% when opening and again when closing; BTC and ETH review rate; other pairs can use different fees and spreads1000x8/10
6Paradex8/100% / 0%50x7.5/10
7Apex Omni8/100.02% / 0.05%100x8/10
8GRVT8/10-0% / 0.045%50x7/10
9DESK8/10-0.01% / 0.017%1000x7.8/10
10Bluefin7.8/100.01% / 0.035%50x8/10
11Lighter7.5/100% / 0%50x7.5/10
12Variational7.5/100% / 0%50x7.5/10

Ranked from the PerpFinder venue registry. Open a venue review for its recorded review date. For order-book venues, the cost comparison tool estimates the trading fee, half-spread, and slippage. Funding costs are separate. Check funding rates for the market and holding period.

How to read the security comparison

This table sorts eligible venues by PerpFinder's recorded editorial security score. The score is a review judgment. It is not an estimate of exploit probability, expected loss, or the chance that a withdrawal will succeed.

The score alone cannot establish which venue is safest. Read the review method, then open the venue profile for its recorded audits, operating history, and source links. Compare the review date with any later changes to the protocol or its controls.

A venue with a high score can still fail. An incident can affect the full deposit, regardless of the trade size. This page does not classify a small deposit as free from material smart-contract risk.

Check what each audit covers

An audit examines a stated scope at a particular time. Check the report's code version, contracts, findings, and unresolved issues. A familiar audit firm or a large report count does not establish that the current deployment received the same review.

A contract audit can leave interfaces, matching systems, price sources, bridges, or administrative controls outside its scope. Read the actual report and the venue's explanation of changes made after the review. Keep unresolved questions separate from confirmed findings.

The GMX review and dYdX review link to venue information and official resources. Use those resources to assess each system; do not assume that a score proves equal coverage across different designs.

Review price inputs and control of funds

Incorrect or delayed price inputs can affect fills and liquidations. Check how a venue obtains prices and what happens when a source stops responding. A security score does not measure current price-feed quality.

Review who can upgrade contracts, pause trading, change risk limits, or control withdrawals. On-chain settlement does not remove every dependency on operators, validators, governance, or external services. The relevant dependencies differ by venue.

Check what happens when matching, a sequencer, a chain, or an interface is unavailable. Determine which actions remain possible and which need an operator. These questions require the venue's current documentation, not an assumption from its chain label.

Compare costs as a separate decision

A lower trading fee does not establish lower security risk. A higher security score does not establish that a venue has enough liquidity for your market or size. Review fees, execution, and funding separately after you understand the system's controls.

The cost comparison tool estimates supported entry execution costs. The funding pages show separate rate observations. Neither calculation measures smart-contract risk or predicts the loss from an incident.