Skip to content

October 3, 2026

SearchAbout
CRYPTO·COINBEAT

Journalism for the digital-asset economy

Ratings / Exchanges & Trading

Best Perpetual DEXs

On-chain perps ranked on the only things that matter in a drawdown: the risk engine, the depth and the funding.

8 services ratedLast verified August 16, 2026Methodology

What the score is made of

Weights total 100% · how each axis is recorded ↓
Risk engine
35%
Liquidation design, oracle sourcing, insurance fund and behaviour in past volatility.
Depth at size
25%
Order-book or pool depth on major markets, and slippage on a realistic clip.
Funding & fees
20%
Taker fees plus the funding rate you actually pay to hold a position.
Decentralisation
20%
Where matching, sequencing and settlement happen, and who controls each.

The table at a glance

8 rated · top score 8.6 · tap a row for the full entry

  1. 01dYdXTraders who prioritise a conservative, well-tested risk engine8.6
  2. 02HyperliquidTraders who want centralised-venue execution without a custodian8.4
  3. 03GMXTraders who want a model they can fully understand8.2
  4. 04AevoTraders who want on-chain options alongside perps8.0
  5. 05VertexArbitrum traders who want a unified book and AMM8.0
  6. 06Gains NetworkTraders who want high leverage on non-crypto markets7.6
  7. 07ApeX ProtocolTraders who want an elastic-margin venue with low fees7.5
  8. 08DriftNobody yet — not until the relaunch lands and users are repaid4.5
Editor’s pickRank 01

dYdX

Traders who prioritise a conservative, well-tested risk engine

The longest-running serious perp venue in DeFi, now on its own Cosmos chain with decentralised validators and an order book that has survived several liquidation cascades intact. Depth has slipped behind newer rivals but the risk design remains the most conservative in the category.

In its favour

  • Longest incident-free record among on-chain perp venues
  • Decentralised validator set running the matching layer
  • Well-documented liquidation and insurance-fund policy

Against it

  • Depth on major pairs now trails the newer leaders
  • Chain migration split liquidity and complicated tooling
  • Launched2017 (v4 chain from 2023)
  • ArchitectureApp-chain, off-chain book, on-chain settlement
  • MarginNon-custodial
Score8.6

The weighted mean of the 4 axes below — each read from the fact printed beside it.

Strongest
Risk engine8.9
Weakest
Funding & fees8.4

Scorecard — and what it was read from

Risk engine
8.9
Operating since 2017 with no loss of user margin on record; v4 publishes its liquidation parameters and maintains an insurance fund with an on-chain balance; liquidations are partial by design.
Depth at size
8.4
Open interest and resting depth sit below the category leader but remain among the deepest on-chain books on major pairs.
Funding & fees
8.4
Published maker/taker schedule with volume tiers; funding published hourly per market.
Decentralisation
8.6
v4 runs as a Cosmos app-chain with an independent validator set operating the matching layer; margin is non-custodial; upgrades go through on-chain governance.

Read the full dYdX review →

The rest of the table

02

Hyperliquid

Traders who want centralised-venue execution without a custodian

A purpose-built L1 running a full on-chain order book with latency close to centralised venues, and it now carries genuine depth on major markets. Validator-set concentration and the youth of the chain are the open questions.

In its favour

  • Fully on-chain order book with centralised-grade execution speed
  • Deepest liquidity of any on-chain perp venue on major pairs
  • No custodial deposit — margin sits in the user's own account

Against it

  • Three market-manipulation episodes during 2025, one leaving ~$4.9m of bad debt in the vault
  • Small validator set that has intervened in live markets by vote
  • Launched2023
  • ArchitecturePurpose-built L1, on-chain order book
  • MarginNon-custodial

Scorecard — and what it was read from

Risk engine
7.6
Runs a fully on-chain order book and liquidation engine with a publicly visible HLP vault. In the March 2025 JELLY episode validators voted within about two minutes to delist the market and settle positions at the attacker's own short price, so HLP closed roughly $0.7m up instead of carrying a ~$13m unrealised loss; delisting votes were moved fully on-chain afterwards. Two further manipulation episodes followed in 2025, the November POPCAT event leaving about $4.9m of bad debt in the vault.
Depth at size
9.5
Highest open interest and resting depth of any on-chain perpetual venue, with every fill settled on its own L1; during the 10 October 2025 liquidation cascade HLP gained roughly $41.5m while more than a thousand accounts were zeroed.
Funding & fees
9.2
Published taker and maker fees at the low end of the category; funding published per market per hour.
Decentralisation
7.6
Matching and settlement run on a purpose-built L1 with a small stake-weighted validator set; margin is non-custodial, and validators have used their discretion to intervene in live markets, which is a governance fact as much as a technical one.

Full Hyperliquid review →

Score8.4
03

GMX

Traders who want a model they can fully understand

A pooled-liquidity design where the counterparty is an on-chain pool with published composition, which makes the risk unusually legible. Zero-slippage execution against an oracle price is elegant until the oracle is the attack surface, as a 2022 exploit demonstrated.

In its favour

  • Simplest counterparty model in the category — the pool is public
  • No price impact on entry for supported markets
  • Long deployment history on Arbitrum and Avalanche

Against it

  • Oracle-priced execution has been exploited on low-liquidity markets
  • Borrow fees make longer-held positions expensive
  • Launched2021
  • ModelPooled liquidity, oracle-priced
  • ChainsArbitrum, Avalanche

Scorecard — and what it was read from

Risk engine
8.4
Pooled counterparty model with oracle-priced execution; the record includes a September 2022 price-manipulation episode on a thin market and a July 2025 re-entrancy in the v1 GLP accounting path that took roughly $40m — the attacker returned the funds and kept a $5m bounty, and v1 was wound down afterwards.
Depth at size
7.8
Fill size is bounded by pool size and published per-market open-interest caps rather than by resting orders.
Funding & fees
7.8
Published open and close fees plus an hourly borrow fee charged against the pool; there is no funding auction between longs and shorts.
Decentralisation
8.8
Contracts on Arbitrum and Avalanche sit behind a governance timelock; execution depends on a designated oracle set; margin is non-custodial.

Full GMX review →

Score8.2
04

Aevo

Traders who want on-chain options alongside perps

One of the few venues where a serious options book sits next to perpetuals on the same margin account, built on a dedicated rollup. Options depth is thin outside the front months, which limits how much of that promise you can use.

In its favour

  • Genuine options and perps on one cross-margined account
  • Dedicated rollup gives predictable fees and fast matching
  • Strong portfolio-margin implementation

Against it

  • Options liquidity thins quickly beyond near-dated strikes
  • Sequencer is operated by the team
  • Launched2023
  • ArchitectureDedicated rollup, off-chain book
  • ProductsPerps and options

Scorecard — and what it was read from

Risk engine
8.2
Options and perpetuals share one portfolio-margin engine with published liquidation parameters and an insurance fund; no loss-of-margin event on record since the 2023 launch.
Depth at size
7.6
Perpetual depth is mid-table; options depth concentrates in near-dated strikes and thins sharply beyond them.
Funding & fees
8.2
Published taker and maker fees; funding published per perpetual market.
Decentralisation
8.0
Runs on a dedicated rollup with a sequencer operated by the team; settlement is on-chain and margin is non-custodial.

Full Aevo review →

Score8.0
05

Vertex

Arbitrum traders who want a unified book and AMM

A hybrid design that pairs an off-chain order book with on-chain AMM liquidity so small orders always fill, deployed across several chains with shared liquidity. Solid engineering that has struggled to attract the depth its design deserves.

In its favour

  • Hybrid book plus AMM removes empty-book failures
  • Cross-chain deployment shares a single liquidity pool
  • Low fees and competitive funding

Against it

  • Depth on major pairs lags the category leaders
  • Sequencing remains team-operated
  • Launched2023
  • ModelHybrid book / AMM
  • ChainsArbitrum and connected networks

Scorecard — and what it was read from

Risk engine
8.0
Hybrid order-book and AMM engine with published liquidation parameters and an insurance fund; no loss-of-margin event on record since the 2023 launch.
Depth at size
7.4
Depth on major pairs trails the category leaders; liquidity is shared across its multi-chain deployments rather than duplicated.
Funding & fees
8.6
Published maker and taker fees among the lowest in the category; funding published per market.
Decentralisation
8.0
Off-chain sequencer operated by the team with on-chain settlement on Arbitrum and connected chains; margin is non-custodial.

Full Vertex review →

Score8.0
06

Gains Network

Traders who want high leverage on non-crypto markets

A synthetic model backed by a single collateral vault that supports forex and equity indices alongside crypto, at leverage no other on-chain venue offers. The vault is both the innovation and the concentration risk.

In its favour

  • Forex, commodities and index markets on-chain
  • Very high leverage available on major pairs
  • Long-running deployment with transparent vault accounting

Against it

  • All counterparty risk concentrates in one collateral vault
  • Synthetic pricing depends entirely on oracle integrity
  • Launched2021
  • ModelSynthetic, vault-backed
  • MarketsCrypto, forex, indices

Scorecard — and what it was read from

Risk engine
7.6
All trades settle against a single collateral vault with published liquidation parameters and spread-based price impact; the vault has run at a temporary deficit during drawdowns but no insolvency event is on record since 2021.
Depth at size
7.0
There is no order book: maximum position size per market is set by published vault parameters rather than by resting depth.
Funding & fees
8.0
Published open and close fees plus a rollover fee charged while a position is held; leverage caps are published per market.
Decentralisation
8.0
Contracts on Arbitrum and Polygon with upgrade rights held under governance; pricing depends entirely on the designated oracle set.

Full Gains Network review →

Score7.6
07

ApeX Protocol

Traders who want an elastic-margin venue with low fees

A competent StarkEx-based venue with cheap execution and a workable cross-margin implementation. It has never attracted enough depth to be a first choice on major markets.

In its favour

  • Low fees and validity-proof settlement
  • Cross-margin account model with no deposit custody
  • Reasonable market coverage for a mid-tier venue

Against it

  • Thin depth outside the largest pairs
  • Operator controls a large part of the stack
  • Launched2021
  • ArchitectureValidity rollup, off-chain book
  • MarginNon-custodial

Scorecard — and what it was read from

Risk engine
7.4
Validity-rollup venue with published liquidation and insurance-fund parameters; no loss-of-margin event on record.
Depth at size
6.8
Depth is thin outside the largest pairs and open interest is a fraction of the category leaders.
Funding & fees
8.4
Published maker and taker fees at the low end of the category; funding published per market.
Decentralisation
7.6
Off-chain matching with validity-proof settlement; margin is non-custodial; the operator controls sequencing and contract upgrades.

Full ApeX Protocol review →

Score7.5
08

Drift

Nobody yet — not until the relaunch lands and users are repaid

Drift was the most complete trading product on Solana until 1 April 2026, when DPRK-linked attackers social-engineered their way to protocol admin and took roughly $285–295m. Repayment runs through recovery tokens backed by a pool of about $148m from Tether and partners, and the relaunch was targeted at the second quarter of 2026.

In its favour

  • Published a detailed incident report and a funded recovery plan
  • Repaid users in full after its earlier 2022 v1 halt
  • Recovery pool backed by Tether and named partners

Against it

  • Roughly $285–295m taken in April 2026; users are not yet made whole
  • Trading suspended pending relaunch, so present depth is zero
  • Launched2021
  • ChainSolana
  • StatusRebuilding after the April 2026 exploit

Scorecard — and what it was read from

Risk engine
3.5
On 1 April 2026 attackers took roughly $285–295m after months of social engineering: DPRK-linked operators posing as a quantitative trading firm got Security Council members to pre-sign dormant transactions using Solana durable nonces, which handed protocol admin to the attacker. The earlier v1 halt in 2022 was settled in full — $19.5m of collateral repaid — but this loss has not been.
Depth at size
4.0
Trading is suspended pending a relaunch targeted at the second quarter of 2026, so there is currently no depth to measure; before April 2026 it held the deepest perpetual liquidity on Solana.
Funding & fees
7.0
Fee and funding schedules remain published, but they price a venue that is not currently operating.
Decentralisation
4.5
The failure was administrative, not a contract bug: pre-signed transactions transferred programme authority to the attacker. The rebuild promises rotated keys and a community multisig, and settlement moves to USDT.

Full Drift review →

Score4.5

↑ Back to the table at a glance

What the record supports

dYdX takes the table on the strength of the longest incident-free record and a conservative liquidation design, even though Hyperliquid still carries far more depth — three manipulation episodes in 2025 and a validator set willing to intervene in live markets is what separates them. GMX remains the simplest counterparty model to reason about. Drift sits last for a reason no design review would have caught: it lost roughly $285–295m in April 2026 to social engineering that ended in a handover of protocol admin, and it has not yet repaid users.

A conclusion drawn from the facts above, and the only part of this page that is.

How a score is read

Each axis is read off the same five bands. They describe what is on the record, not how impressed we are.

9.0–10
Documented and independently verifiable
The claim is evidenced by a published record a third party can check — an attestation, an on-chain contract, a regulator's register — and nothing adverse is on file.
8.0–8.9
Documented, with gaps
Evidence exists but is partial, dated, or covers only part of what the axis measures.
7.0–7.9
Self-reported only
The operator publishes the information and no independent party has verified it.
6.0–6.9
Adverse event on record
A recorded incident, enforcement action or failure that has since been resolved, remediated or repaid.
Below 6
Undocumented or unresolved
No published evidence, or an incident with no resolution on the record. An absence of evidence is scored as an absence.

How we scored this table

Perpetual venues differ least in calm markets and most in the minutes that decide whether a liquidation cascade stops. This table therefore records mechanism facts: the mark-price source and its update interval, whether liquidations are partial or full, the insurance fund's on-chain balance, and whether profitable traders can be auto-deleveraged.

Recorded volatility events count for more than documentation. Where a venue has been through a cascade, an oracle failure or an intervention, the event and its outcome are on the page, and the risk-engine score is read from that record rather than from the design description.

Depth is recorded as open interest and price impact for a fixed clip, checked against on-chain settlement where the venue settles on-chain. Decentralisation is recorded as who operates matching, who runs the validator set or sequencer, and who can halt the system — not as a label.

  • Every score on this page carries the fact it was read from, printed beside the bar.
  • Funding is recorded as an average cost of carry, not a snapshot rate.
  • Volume that cannot be reconciled with on-chain settlement is not recorded.

What each axis records, and where the facts come from

Risk engine35%
How the mark price is sourced and how often it updates; whether liquidations are partial or full; the insurance fund's on-chain balance and how it is funded; whether profitable traders can be auto-deleveraged; what the venue did during recorded volatility events.
Source: Protocol documentation, on-chain insurance-fund balances, liquidation and settlement records, published post-event reports.
Depth at size25%
Resting depth and price impact for a fixed clip on major markets; open interest per market; whether reported volume reconciles with on-chain settlement.
Source: Order-book and pool snapshots, on-chain settlement records, published open-interest data.
Funding & fees20%
The published taker and maker schedule; the funding formula, its interval, and the recorded average cost of carry on major markets.
Source: Fee documentation, historical funding records per market.
Decentralisation20%
Where orders are matched, who operates the sequencer or validator set, whether margin is custodial, and who can upgrade or halt the venue.
Source: Chain and validator documentation, contract admin addresses, published upgrade and intervention processes.

Frequently asked questions

What is the biggest risk specific to perpetual DEXs?+

Oracle and liquidation design. If the mark price can be pushed or lags a fast move, positions get liquidated at prices that never existed on a real book — and on some venues profitable traders absorb the shortfall through auto-deleveraging.

How does funding actually cost me money?+

Funding is paid periodically between longs and shorts to keep the perpetual price near spot. Holding the crowded side of a trending market can cost more over a few weeks than the entire fee schedule, which is why cost of carry belongs in any comparison.

Are perp DEXs really decentralised?+

Partly. Most run matching off-chain for latency and settle on-chain, so you get non-custodial margin with a centralised sequencer. That is a meaningful improvement over full custody, but it is not censorship-resistant execution.

Should beginners trade perpetuals at all?+

No. Leverage plus funding plus liquidation mechanics is three unfamiliar systems interacting at once, and the failure mode is a zeroed account rather than a bad entry. Learn spot first; perps are not a way to trade smaller.

More in Exchanges & Trading