How to Compare Crypto Exchanges: Licences, Fees and Fit
A comparison table feels like evidence. Often it is a ranking of numbers that were defined differently, measured on different dates, and weighted by someone whose priorities were never stated. This guide sets out the three ways those tables mislead, then gives a comparison method that survives being checked — starting with the dimensions where a bad outcome is hardest to reverse.
Cover artwork: CoinSoso · AI-generated, not a photograph of a real venue, product or market capture.
Key takeaways
- A comparison is only as good as its definitions: registered users, monthly transacting users and account counts measure different things and cannot be ranked against each other.
- Coinbase's own filing defines its headline figure as a rolling 28-day transacting count and warns it may overstate unique customers — it is not a registered-user number.
- Even a single venue's own pages can disagree on the same metric on the same day, which is why every cell needs a date and a source.
- A ranking is a weighting in disguise; if the weights are not published, the ordering is the author's preference rather than a finding.
- Compare in an order set by recoverability: custody and withdrawal first, then jurisdiction, then cost, because a cheap venue you cannot exit is not cheap.
Three ways a comparison table misleads
A comparison is an argument wearing the clothes of a fact. That is not a criticism of all comparisons — some are careful, sourced and dated. It is a warning that the format hides three decisions, and each of them can flip the answer without anything on the page looking wrong.
Definitions. Two numbers in the same column are only comparable if they measure the same thing.
Dates. Rates, licences, product lines and minimums change. A table without a date is a claim about an unspecified moment.
Weights. A ranking is a weighted average. If the weights are not published, the ordering reflects the author’s priorities, not the reader’s.
The rest of this guide takes each in turn and then sets out an order for the comparison itself.
The definition problem, in the wild
The clearest example is the number every exchange marketing page leads with.
| Source | The figure it publishes | What that figure actually counts |
|---|---|---|
| Binance | 323M+ registered users | Accounts created |
| OKX | 120M+ users globally | Users, as the page defines it |
| Bybit | 80M+ registered users | Accounts created |
| Coinbase | 9.2M monthly transacting users (FY2025 average) | Consumers transacting in a rolling 28-day window |
The first, second and third are marketing figures of different vintages and scope. The fourth is a metric a public company is required to define, and defines it precisely:
“We define a Monthly Transacting User (‘MTU’) as a consumer who actively or passively transacts in one or more products on our platform at least once during the rolling 28-day period ending on the date of measurement.”
The same filing then does something marketing pages do not: it warns about its own number. Coinbase states that MTUs “may overstate the number of unique customers who have registered an account on our platform as one customer may register for, and use, multiple accounts.”
So a table that lists 323M, 120M, 80M and 50M for Coinbase is not comparing four venues on one measure. It is comparing three unverified marketing counts with a number that means something else entirely, and the Coinbase entry is the one most likely to be wrong in a specific, checkable way: its own disclosure says the closest official figure is 9.2 million MTUs, and that even that is not a count of registered customers.
This matters beyond vanity metrics, because the same pattern recurs on every dimension. “Maker fee” can mean the base tier or the top tier. “Licensed” can mean the operating entity or an affiliate. “Cold storage” can mean most of the assets or a portion of them. Whenever two cells sit side by side, the question is not what the number is but what it measures.
The date problem
The second failure is subtler because it produces a table that is internally consistent and still wrong.
Venues do not update all their pages at once. OKX, read on a single day, publishes a 103% reserve ratio with 46 monthly audits and $25.7 billion in primary assets on one page, and a 48th proof-of-reserves report with $29.4 billion in primary assets on another. Both are the venue’s own material, both are live, and they describe different states of the world.
A comparison that took the 103% from the first page and the $25.7 billion from the same page would be internally coherent and a month out of date, with no indication of that anywhere in the table — which is why every cell in a comparison needs a date attached to it, not a date in the footer.
The practical rule:
- Date the source, not the reading. “Read on 2026-10-18” is useful; “data as of August 2026” is what you actually need. When a page gives both, both belong in the cell.
- Prefer the specific document to the summary page. A dated report beats a marketing page; a filing beats both.
- Re-read before acting. A saved screenshot has no expiry printed on it.
The OKX reserve report guide works through that particular discrepancy in detail, and it is worth reading as a worked example of how much a dated primary document gives you that a summary does not.
The weighting problem
Suppose a table scores two venues out of ten on three dimensions. The composite will rank them, and the ranking will be arithmetically correct and analytically meaningless, because the weights decide the winner.
| Dimension | Venue A | Venue B |
|---|---|---|
| Cost at your volume | 9 | 6 |
| Regulatory transparency | 4 | 8 |
| Asset coverage you need | 7 | 7 |
Equal weights give A a composite of 6.67 and B a 7.00 — B wins. Weight cost at 60% and transparency at 20% and A wins at 7.60 to 6.60. Same facts, opposite conclusion, and a reader who is not shown the weights cannot know which ranking they were given.
There is a second problem with composites beyond the weights: they average across dimensions where the outcomes are not substitutable. Saving 0.02% on fees does not compensate for a venue you cannot withdraw from. Averaging those two into one score implies a trade-off that most people would not accept if it were stated.
An order for the comparison
Comparisons should run in order of how hard a bad answer is to recover from, not in order of how easy it is to look up. Fees are the easiest thing to compare and the least consequential when something goes wrong.
1. Custody and exit. Can you withdraw the asset you hold, on the network you hold it on, without a support ticket? What does the venue publish about backing customer assets, when was it published, and does it distinguish exchange-held from third-party custody? The OKX proof-of-reserves guide is a template for reading that material, whichever venue published it.
2. Jurisdiction and entity. Which legal entity holds your account, where is it registered, and what does that regulator actually enforce? This is the dimension where badges and reality diverge most, and how to check an exchange’s regulatory claims sets out how to match the entity in the terms to a public register.
3. Access. Can you complete verification, deposit and withdraw in your country, in your currency, through your bank? An unusable venue is not a candidate regardless of its fee table.
4. Asset and product coverage. Does it list the assets you actually want, with the order types you need? Check the symbol’s filters rather than the marketing page — a pair with a minimum notional above your intended order size is not available to you in practice.
5. Cost. Maker and taker at your own tier, on your own pair group, plus the spread you would cross, plus deposit and withdrawal charges, plus funding if the product is a perpetual. The fee schedule guide covers how the tiers and windows decide which row applies.
6. Liquidity and depth. A fee advantage on a pair whose book is two basis points deep is not an advantage. Depth around the size you trade is the usable measure.
7. Operations. Withdrawal limits, processing times, support responsiveness, scheduled maintenance. Unglamorous, and the thing you notice on a bad day.
Building a comparison you can defend
A comparison worth acting on looks like a spreadsheet with more columns than a marketing page, not fewer.
- One row per dimension, one column per venue, and a source plus a date in every cell. A cell without a source is a memory.
- Record the measurement, not the label. “0.1% maker, base tier, spot crypto pairs, read 2026-10-18” is a fact; “low fees” is not.
- Keep a separate note for what the venue would not tell you. An unverifiable claim is information about the venue, and it should be visible in the table rather than averaged away.
- State your own weights, in writing, before you look at the scores. If the ranking surprises you, the weights were not what you thought they were.
- Never compare a claim against a different claim’s number. Compare a measurement against the same measurement.
There is one more thing to be honest about, and it applies to this site as much as any other: exchange comparison content is frequently monetised through affiliate arrangements, which creates an incentive to rank the venues that pay. CoinSoso carries exchange profile pages and affiliate links, and gates registration links from a configuration file rather than inserting them per article. Treat the ordering in any comparison — including ours — as a claim that needs its sources checked, not as a finding.
What a good comparison will not tell you
It will not tell you which venue is safe, because safety is not one property. It will not tell you which venue will be cheapest next quarter, because rates change and the venue typically gives about a week’s notice. And it will not tell you whether you should be trading at all.
What it can do is make the trade-offs visible, which is the only thing a comparison was ever entitled to do.
The narrower versions of the same research are in how to read an exchange fee schedule, how to check an exchange’s regulatory claims, and — for the case where the question is not which exchange, but whether an exchange at all — centralised versus decentralised venues. The trading fee calculator prices the cost dimension once you have your own numbers.
Risk note. The Coinbase definition and caveat are quoted from its Form 10-K for the year ended 2025-12-31, filed 2026-02-12 and read on 2026-10-18; the Coinbase monthly figure is a rolling 28-day transacting count and is not a registered-user number. The user figures attributed to Binance, OKX and Bybit are the figures those venues publish on their own pages and are repeated here as published labels, not as verified measurements. The OKX figures are quoted from the two OKX pages named in the sources, read on 2026-10-18, precisely because they disagree. The scores in the weighting example are illustrative and are not a ranking of any venue. Venue terms, fees and product availability change; re-check the primary sources rather than this article. Nothing here is investment advice or a recommendation to use any platform.
Sources
How this was researched. Written from primary documentation and the exchange or protocol's own published material, with sources listed at the end of the article.
Primary sources for the factual claims on this page. A date appears only where a human has actually read the source; where it is missing, the check is still outstanding.
Frequently asked questions
Which crypto exchange is the safest?
That question cannot be answered from a comparison table, because 'safe' bundles several different risks that move independently. A venue can hold customer assets in full and operate under a licence that gives customers no practical recourse; another can hold assets at third-party custodians and be answerable to a regulator that publishes enforcement. The useful version of the question is narrower: what specifically are you worried about, and what published evidence bears on it?
Why do different sites give different user counts for the same exchange?
Because the exchanges define the metric differently and rarely publish the same one. Binance, OKX and Bybit report large user figures on their own marketing pages, while Coinbase reports Monthly Transacting Users — 9.2 million for FY2025 — defined as consumers transacting in a rolling 28-day window, with a published warning that the figure may overstate unique customers. These are three different measurements, and placing them in one column implies an equivalence that does not exist.
How do I compare fees fairly?
Compare the same pair, the same tier, the same order type and the same date. Then add what the table omits: the spread you would cross, the deposit and withdrawal charges, and for derivatives the funding rate. A maker rate on a pair you cannot get depth on is not a real cost advantage, and a taker rate that comes with a wide spread is understated.
What does an exchange licence actually give me?
It depends on the entity, the regulator and the permissions. A licence held by a European subsidiary does not necessarily cover a customer in another region, and the legal entity named in the terms is the one that matters. Checking it means matching the entity in the terms of service to an entry in a public register, which is a different exercise from reading a badge on a homepage; the method is set out in the linked regulatory-claims guide.
Should I use a comparison table with a score out of ten?
Only if it publishes its weights and its sources with dates. A composite score hides the trade-offs that a comparison is supposed to expose: two venues can score identically while being worse than each other on every dimension the reader cares about. A table of raw, dated facts that you weight yourself will be less convenient and more useful.
Where to go next
Independent research. Each card shows the platform’s invite code, and its Create Account button is a referral link that may earn us a commission at no extra cost to you. Whether an account can be opened is decided by the platform at sign-up.
Binance
One of the largest global exchanges by traded volume, with a very broad product range.
20% Trading Fee Rebate
Use invite code con888 when signing up to enjoy trading fee benefits.
Download from the exchange's official download service. Verify the source and follow the on-screen installation instructions.
Do not install an APK from an unknown source, or one that has been modified or cannot be checked.
OKX
A large global exchange offering spot, derivatives and a self-custody wallet.
20% Trading Fee Rebate
Enter invite code con888 when signing up to check available trading fee benefits.
Download from the exchange's official download service. Verify the source and follow the on-screen installation instructions.
Do not install an APK from an unknown source, or one that has been modified or cannot be checked.