How to Compare DEXs on What Actually Matters

Choosing a decentralised exchange on its swap fee is like choosing a bank on its logo. The fee is one number among five that matter, and it is the only one every venue advertises. The others — what licence the code carries, whether the contracts can be changed after deployment, which networks the protocol actually claims, and whether audits name their auditors — are all published, and they diverge far more sharply than the fee does.

This is the reasoning behind our DEX ranking, written so you can apply it to any protocol.

What is the fee, and who splits it?

Swap fees in this category no longer come as a single number. Three models coexist:

• A flat rate across all pools — the original design, simple to compare and increasingly rare.
• Discrete tiers, typically running from a hundredth of a percent for stable pairs up to one percent or more for volatile ones, chosen when the pool is created. Most major venues now work this way, and the tier you pay depends on which pool routes your trade rather than on the protocol.
• Fully settable fees, where the pool creator picks any value inside a range — in the widest published case, anywhere from zero to one hundred percent. That is not a hypothetical: an architecture that permits it puts the burden on the router and on you to check what a specific pool charges.

Then there is the split. The headline fee is divided between liquidity providers and the protocol, and the protocol's share is a separate published parameter: a third of the LP fee capped at a stated maximum in one design, half the trade fee in another, or an explicit breakdown to LPs, treasury and other funds. As a trader the split does not change what you pay. It tells you how the protocol is funded and how much it can raise its take without touching the advertised rate.

A fourth model deserves its own warning: venues that do not state fees as percentages at all, describing instead a surplus retained by an API provider or fees "covered by the taker". Aggregators frequently work this way, and integrators can add their own fee on top through a parameter in the routing call. The cost is real; it is simply not quotable in advance, which is a genuine difference from a published tier.

Does a public repository mean open source?

No, and in this category the gap between the two is wide enough to be the single most differentiating criterion we score.

What you find when you read the actual LICENSE file, rather than the documentation's description of it:

Licences read from the repositories themselves and fees from each protocol's own documentation, 24 July 2026. Licence terms and change dates are stated by the projects; each is cited on the review it came from.

Four distinct positions appear in that column, and they are not interchangeable. A recognised open-source licence grants you rights permanently. A delayed-open licence grants them on a stated future date — and note that some of those dates have already passed, converting the code automatically, which is a fact worth checking rather than assuming. A revocable licence grants rights the licensor may withdraw. And code published with no licence, or with an explicit statement that no rights are granted, reserves everything by default no matter how readable it is.

Why it matters practically: the licence decides who may fork the code, audit it at depth, or ship a fixed version if the team will not. In a category whose entire pitch is that you are trusting code rather than a company, that is not a legal footnote.

Can the contracts be changed after deployment?

"Decentralised" says nothing about upgradeability. Three arrangements exist and each has a different failure mode:

• Immutable core. The deployed contracts cannot be altered, paused or reversed by anyone, including the team. Maximum credible neutrality, and no emergency stop when something goes wrong.
• Immutable pools, upgradeable factories. Existing pools keep their rules; new deployments can change. Users migrate by choice rather than by decree — a middle position several protocols state explicitly.
• Upgradeable contracts. An owner can change behaviour, usually behind a timelock. This is a trust relationship with whoever holds that key, and it should be described as one.

The best signal here is not which model a protocol picked but whether it states the model plainly, including which parts remain upgradeable. Protocols that publish "core not upgradeable, cannot be paused or reversed" or "protocol immutable, factories upgradable" have given you something to verify. Silence usually means the third option.

Are you trading on the venue you clicked?

Increasingly, no. A great deal of on-chain volume now arrives through aggregators and routers that split a single trade across several pools, sometimes across several protocols, choosing the path that returns the most output after gas. The interface you used may be a front-end to a routing engine rather than to a specific pool.

That changes what "the fee" means. Your cost becomes the sum of whatever fee each pool in the route charges, plus the price impact of each hop, plus any fee the router or its integrator adds — and the routing fee is frequently set by the integrator, not the protocol, through a parameter in the swap call. Two front-ends to the same underlying liquidity can therefore charge you different amounts for an identical trade.

What to check, in order of how much it usually costs: whether the interface shows you the route it chose and the fee tier of each hop; whether it displays a fee line separate from price impact, since a single "estimated cost" figure conceals which of the two you are paying; and whether the quoted output is after all fees. A venue publishing a clear tier and showing its route is easier to price than an aggregator quoting one net number, even if the aggregator's number turns out better.

What does a fee tier buy a liquidity provider?

The fee is a trader's cost and a liquidity provider's revenue, and the tier structure exists because those two roles want opposite things. Stable pairs trade at a hundredth of a percent because the pool bears almost no divergence risk; volatile pairs need one percent or more to compensate for it. A tier is therefore a rough price for the risk of holding both sides of that pair.

Concentrated liquidity changed this trade. Rather than spreading capital across all prices, a provider selects a range, earning far more fees per unit of capital while it holds — and earning nothing at all once price leaves the range. This turns liquidity provision from something close to passive into a position that needs management, and it is the main reason headline LP yields in this category can look extraordinary while realised returns disappoint.

Two consequences worth carrying into any comparison. For traders, deep concentrated liquidity near the current price means excellent execution for ordinary sizes and a sharper deterioration for large ones once the range is exhausted — depth is no longer uniform across prices. For providers, an advertised fee tier is not a yield: it is a rate applied only to the volume that trades inside your chosen range, minus what the position lost by rebalancing into the falling asset along the way.

Do the chain and audit claims survive checking?

Two criteria that are quick to check and surprisingly often fail.

Chain coverage should come from the protocol's own deployment documentation — a named list of networks with addresses — rather than an aggregator's count. Published lists in our sample ranged from a single network to thirty-eight, and one protocol's own pages could not produce a verifiable list at all, which we score as the absence of a published answer rather than guessing at a number.

Audit disclosure means reports published with the firms named and, ideally, dated. Several of the largest venues in this category name multiple auditors with dates; several others say nothing about audits anywhere on their own pages. An unnamed audit cannot be checked, and no audit page at all is scored as the absence it is — not as evidence of insecurity, but as an absence of evidence, which for a contract holding pooled funds is a real finding.

What this ranking does not tell you

Trading experience is not scored: routing quality, MEV exposure and price impact are properties of a specific trade at a specific moment, not of a protocol, and we could not measure them to the same standard across venues. Nor do we score past incidents unless a protocol documents one itself — reporting an exploit from memory, on a page about somebody else's security, is not something we are willing to do.

What the ranking does tell you is what a protocol has committed to in writing: what it charges, what rights its code grants, whether it can change the rules, and who has checked it. Those four answers are stable, checkable, and enough to separate protocols that most comparisons treat as interchangeable.

Sources

• Fee models, licences, upgradeability statements, chain lists and audit disclosures as published in each protocol's own documentation and repositories, cited individually on the reviews linked above, read 24 July 2026
• Our scoring rubric for this category — Ratings Methodology

How automated market makers work mechanically is covered in our guide to automated market makers. This article is about choosing between the venues that run them.