The best crypto exchanges, ranked on what a trade costs and what backs your balance
We didn’t read the fee pages. We walked six live order books and priced the same $10,000 trade on each — the gap between cheapest and dearest is more than six-fold. Then we checked every proof-of-reserves, licence and enforcement record behind them. The cheapest venue is not the winner, and the biggest hack in history did not disqualify anyone.
The verdict, in three lines
- Best overall
- Kraken 7.8/10
- Cheapest to trade
- OKX 0.1% on $10k
- Weakest evidence your money is there
- BitMart 1/10 custody
Custody carries 40% of the score, so the cheapest venue does not win. Why we weight it that way.
⚠ The cheapest venue may not be open to you — check the availability column before the price.
Which one should you use?
The table below has every number behind these answers. If you just want the short version, find your situation.
It has the strongest evidence that customer money is really there, works in every state except Maine and New York, and its fees are published upfront. You pay about $26 on a $10,000 trade — roughly $16 more than the cheapest venue, which is the price of the safest custody record in the table.
Both are licensed to serve you and both can prove their reserves. Bitvavo is the largest euro venue and keeps customer assets in a separate legal entity, so the company’s own finances cannot touch them. Binance is cheaper and has suspended most EU services since July 2026.
About $10 on a $10,000 trade, the joint-cheapest we measured. The trade-off is custody evidence nobody outside the company signs off, and a 2025 guilty plea in a US federal court. Cheap is a real advantage; it is not the same as safe.
Every venue in this table is a company holding your money. For anything you are not actively trading, a hardware wallet removes that company from the equation entirely — our top pick costs less than the fees on a single large trade.
All 29 exchanges: what it costs, and where you can actually use it
cost measured 14 Jul 2026, 12:00 UTC| # | Exchange | Our scoreout of 10 | What a $10,000 trade costsfee + spread, measured | Can you check your money is there?who verifies it | Can you open an account?US · EU · UK |
|---|---|---|---|---|---|
| 1 | Kraken | 7.8 | $260.26% of the trade | Yes — an outside firm checks | US some statesEU ✓UK ✓ |
| 2 | OKX | 7.3 | $100.1% of the trade | Only their own word | US some statesEU ✓UK some states |
| 3 | MEXC | 7.2 | $60.0565% of the trade | Yes — an outside firm checks | US ✕EU ✕UK ✕ |
| 4 | Bitvavo | 7.1 | not measuredHTTP 403 | Yes — an outside firm checks | US ✕EU ✓UK ✕ |
| 5 | Bybit | 7.1 | $100.1% of the trade | Yes — an outside firm checks | US ✕EU some statesUK some statesYes — confirmed |
| 6 | Upbit | 7.0 | $50.05% of the trade | Yes — an outside firm checks | US ✕EU ✕UK ? |
| 7 | Binance | 6.8 | $100.1% of the trade | Only their own word | US some statesEU ✕UK ✕ |
| 8 | Bitget | 6.8 | $100.1% of the trade | A page with no numbers in it | US ✕EU ✕UK ? |
| 9 | Coinbase | 6.6 | $600.6001% of the trade | No — nothing to check | US ✓EU ✓UK ✓ |
| 10 | WhiteBIT | 6.5 | $100.1045% of the trade | Out of date | US ✕EU ✓UK ? |
| 11 | Bitstamp | 6.5 | $300.3% of the trade | No — nothing to check | US ✓EU ✓UK ✓ |
| 12 | Gemini | 6.3 | not measuredHTTP 451 | Partly — audited accounts only | US ✓EU leftUK left |
| 13 | BingX | 6.3 | $100.1032% of the trade | Only their own word | US ✕EU ✕UK ✕Promised, unconfirmed |
| 14 | Crypto.com | 5.9 | $250.25% of the trade | Out of date | US some statesEU ✓UK ✓ |
| 15 | Phemex | 5.5 | $100.1001% of the trade | Only their own word | US ✕EU ✕UK ✕Promised, unconfirmed |
| 16 | KuCoin | 5.4 | $100.1% of the trade | Only their own word | US ✕EU ✕UK ✕ |
| 17 | Bitso | 5.4 | $660.6574% of the trade | Only their own word | US ?EU ✕UK ? |
| 18 | Gate | 5.1 | $210.2063% of the trade | Only their own word | US some statesEU ✓UK ✕Promised, unconfirmed |
| 19 | Bitrue | 5.0 | $100.098% of the trade | A page with no numbers in it | US ✕EU ✕UK ?Yes — confirmed |
| 20 | DigiFinex | 4.9 | $200.2% of the trade | Only their own word | US ✕EU ✕UK ✕ |
| 21 | HTX | 4.8 | $210.2062% of the trade | Only their own word | US ✕EU ✕UK ✕ |
| 22 | Bitfinex | 4.5 | $200.2% of the trade | Out of date | US ✕EU ?UK ✕ |
| 23 | Poloniex | 4.4 | $240.2412% of the trade | Only their own word | US ✕EU ✕UK ✕Promised, unconfirmed |
| 24 | Independent Reserve | 4.3 | $900.8967% of the trade | Partly — audited accounts only | US ✕EU ?UK ? |
| 25 | XT.com | 4.2 | $200.2% of the trade | A page with no numbers in it | US ✕EU ✕UK ✕Promised, unconfirmed |
| 26 | CoinEx | 3.9 | $230.2327% of the trade | Only their own word | US ✕EU ✕UK ✕Promised, unconfirmed |
| 27 | LBank | 3.8 | $100.1% of the trade | A page with no numbers in it | US ✕EU ✕UK ? |
| 28 | BitMart | 3.2 | $250.25% of the trade | No — nothing to check | US some statesEU ✕UK ✕No — still unpaid |
| — | AscendEXDEFUNCT | — | not measuredUnexpected token '<', "<!doc | not yet researched | US —EU —UK —Promised, unconfirmed |
Cost and depth are measured by us from each venue’s live public order book; taker fees are the venues’ own published rates. Availability is sourced to each company’s own terms or a regulator’s register — hover any cell for the detail.
Read the cost column carefully. We measure the global order book. Where a company serves the US through a separate arm (Binance.US, OKX US, Gate US), that platform has its own, usually thinner book — the number here is not the one an American would get.
The cost column is what a $10,000 market buy of bitcoin actually cost us on that venue — the trading fee plus the gap between the price shown and the price you get. We walked each order book ourselves rather than reading the fee page. Click any exchange for the full reasoning behind its score.
⚠ AscendEX has shut down
CEASED OPERATIONS on 1 July 2026 after failing to obtain a MiCA licence, amid a liquidity crisis. Withdrawals were moved to manual review with no guarantee of processing. Our own measurement run could not reach its order book — the API returns a web page. Do not sign up. We list it because a comparison that quietly deletes the venue that just collapsed teaches the reader nothing.
This is the argument for measuring rather than compiling. Our automated run flagged the venue because its order book endpoint stopped returning an order book — we did not learn it from a press release. A ranking refreshed once a quarter would still be recommending it. How we measure
HOW WE SCORED THIS+
Every criterion carries a published weight and the weights sum to 100. Custody takes 40% of it, deliberately. Anything we cannot verify is not listed below at all — we delete the criterion instead of carrying it unscored. Full scale and universal criteria are in the methodology.
40%Custody of funds+
Proof-of-reserves: does it exist, who attests it, does it include liabilities (a PoR without liabilities is close to meaningless), how often. Insurance fund: size, is the address public, and has it ever actually paid out. Hack history — and above all whether users were made whole. Licences by jurisdiction, and the enforcement record behind them. Every claim here is sourced to a primary document: a regulator’s register, a court filing, or the venue’s own published proof.
16%True cost of trading+
Measured by us, not read off a fee page. We walk the live order book and fill $1,000 / $10,000 / $100,000 of BTC/USDT, then add the venue’s published taker fee. That sum — fee + spread + realised slippage — is the effective cost. Separately: the markup on network withdrawal fees, and the hidden spread inside “convert” / “buy in one click” flows.
11%Liquidity and execution+
Order-book depth within ±0.1% and ±0.5% of mid, measured from the live book. Uptime over 12 months from the status page. Behaviour under stress: were there outages on days the market moved more than 5%.
9%Assets and networks+
Number of pairs, adjusted for quality: the share of listings delisted within a year. Which networks are actually supported for deposit and withdrawal — not merely “supports USDT”.
11%Fiat on/off ramp+
Which methods are actually available per country, their fees, and their limits per verification tier — read from the venue’s published terms and cited. Where a venue does not publish this, that opacity is itself the finding, and it scores accordingly.
8%KYC and access+
Verification tiers and their limits, what is genuinely usable without verification, and the country deny-list — from the venue’s published policy, cited. A venue that will not tell you your withdrawal ceiling before you sign up is marked down for it.
5%Product, API and docs+
API documentation quality, whether rate limits are documented, whether a public status page with uptime history exists, and the API response latency we measured ourselves while pulling the order book.
Every criterion above is verified — measured by us, or sourced to a primary document. There is no placeholder criterion and no asterisk: the score rests on 100% of the published methodology. If we cannot verify something, we delete the criterion rather than carry it unscored. That is why there is no “support” score here: judging it honestly means opening real tickets and timing the replies, and we have not done that. It returns when we do — with a changelog entry saying so.
Every exchange, one page each
One profile per company — not per legal entity. Each page carries our measurements, the per-criterion scorecard, the proof-of-reserves position, regional access with sources, and the incident record with the only question that decides it: did users get their money back?
WHO WE EXCLUDED, AND WHY+
A rating that contains only winners is a catalogue. These are the hard gates: fail one and no fee, volume or score gets you into the table.
Global disqualifier. If a custodial business will neither show its reserves nor answer to a regulator, there is nothing to rate — you are trusting a promise. We do not rank promises.
We size real orders against the live book. Where the depth cannot support the volume a venue reports, the volume figure is doing marketing, not reporting. Excluded rather than rated.
Global disqualifier, regardless of fees or liquidity. Note how this cuts: Bybit lost the largest sum in exchange history and is still in the table, because users were made whole and withdrawals never stopped. The test is not whether something went wrong. It is what happened to your money afterwards.
- × No proof-of-reserves and no licence.
- × Funds not returned to users after a hack.
- × A pattern of complaints about blocked withdrawals.
- × Confirmed wash trading.
OUR RAW MEASUREMENTS+
The full capture behind the cost column. One timestamp, six venues, the same pair and order sizes — reproducible by anyone willing to run the same walk against the same public endpoints.
| VENUE | BEST ASK | SPREAD | TAKER | SLIP $1K | SLIP $10K | SLIP $100K | API LATENCY |
|---|---|---|---|---|---|---|---|
| Binance | 62,860.73 | 0% | 0.1% | 0% | 0% | 0 | 460 ms |
| Coinbase | 62,789.62 | 0% | 0.6% | 0% | 0.0001% | 0.0039 | 183 ms |
| Kraken | 62,793.5 | 0.0002% | 0.26% | 0% | 0% | 0.001 | 224 ms |
| OKX | 62,864.8 | 0.0002% | 0.1% | 0% | 0% | 0.0005 | 330 ms |
| Bybit | 62,867.2 | 0.0002% | 0.1% | 0% | 0% | 0.0043 | 254 ms |
| Gate.io | 62,843.1 | 0.0002% | 0.2% | 0% | 0.0063% | 0.0206 | 3028 ms |
| KuCoin | 62,858.7 | 0.0002% | 0.1% | 0% | 0% | 0.0036 | 465 ms |
| Bitget | 62,860 | 0% | 0.1% | 0% | 0% | 0.0081 | 521 ms |
| MEXC | 62,857.57 | 0% | 0.05% | 0.0003% | 0.0065% | 0.009 | 456 ms |
| HTX | 62,859.8 | 0% | 0.2% | 0% | 0.0062% | 0.0165 | 468 ms |
| Crypto.com | 62,803.5 | 0% | 0.25% | 0% | 0% | 0.0023 | 572 ms |
| Bitfinex | 62,900 | 0.0159% | 0.2% | 0% | 0% | 0.0162 | 350 ms |
| Bitstamp | 62,785.09 | 0% | 0.3% | 0% | 0% | 0.0083 | 320 ms |
| BingX | 62,860.73 | 0% | 0.1% | 0.002% | 0.0032% | 0.0061 | 460 ms |
| BitMart | 62,860.74 | 0% | 0.25% | 0% | 0% | 0 | 631 ms |
| LBank | 62,861.02 | 0% | 0.1% | 0% | 0% | 0 | 1067 ms |
| Phemex | 62,865.23 | 0.0263% | 0.1% | 0% | 0.0001% | 0.0185 | 911 ms |
| WhiteBIT | 62,863.78 | 0% | 0.1% | 0% | 0.0045% | 0.0181 | 364 ms |
| CoinEx | 62,846 | 0.0016% | 0.2% | 0.0098% | 0.0327% | 0.0508 | 595 ms |
| Poloniex | 62,861.28 | 0.0144% | 0.2% | 0.0336% | 0.0412% | 0.0539 | 588 ms |
| Bitrue | 62,860.73 | 0% | 0.098% | 0% | 0% | 0.0031 | 1050 ms |
| DigiFinex | 62,860.73 | 0% | 0.2% | 0% | 0% | 0 | 668 ms |
| XT.com | 62,860.73 | 0% | 0.2% | 0% | 0% | 0 | 542 ms |
| Upbit | 92,960,000 | 0.0097% | 0.05% | 0% | 0% | 0 | 2390 ms |
| Bitso | 62,809 | 0.0318% | 0.65% | 0.002% | 0.0074% | 0.0093 | 600 ms |
| Independent Reserve | 62,786.97 | 0.0366% | 0.5% | 0% | 0.3967% | 0.5799 | 2627 ms |
Captured 14 Jul 2026, 12:00 UTC · pair · ChainWatch Daily automated Level-A runner
Spread, depth and slippage are measured by us from each venue’s live public order book at measuredAt. Taker fees are the venues’ own published entry-tier rates (sourced, not measured). Effective cost combines the two. Venues that could not be measured are listed with the reason rather than dropped. No field test (deposit/withdraw/support) is included — that is Level B and requires a funded account.
WHAT CHANGED SINCE THE LAST UPDATE+
Methodology changes are published, not applied quietly. A ranking that reshuffles without saying why is just an opinion with a fresh date on it.
The “support” criterion was REMOVED, not left unscored. Judging support honestly means opening real tickets as a paying customer and timing the replies; we have not done that. Rather than carry a 5% placeholder marked n/a — which quietly dilutes every other weight and lets a score hide behind an asterisk — we deleted the criterion and redistributed its weight across criteria we can actually verify. Scores now rest on 100% of the published methodology. If we later run the tickets, the criterion comes back and this changelog will say so on the day it does.
Custody weight raised from 25% to 40%, making it the dominant criterion. Cost fell to 16%, liquidity to 11%, assets to 9%, UX to 5%. Reason: an exchange is a custodian before it is a marketplace. A 0.5% cost difference is worth $50 on a $10,000 trade; a custody failure is worth $10,000. Weighting them close together implied a trade-off that does not exist. The reshuffle moved Kraken to first and dropped Binance from second to fourth — the ranking now rewards what protects your balance rather than what shaves basis points off a fill.
Custody, fiat, KYC and product criteria closed with sourced public-record research: proof-of-reserves attestation (does it include liabilities? is there a named third-party auditor?), licence registers, enforcement records, and — the question that matters most — whether users got their money back after each incident. Every claim about a named company now carries a link to a primary document.
First publication. Effective trading cost measured by walking each venue’s live order book at $1k / $10k / $100k, rather than quoting fee schedules.
Do you need a centralised exchange at all?
For buying and holding, you need an exchange once — to convert fiat into crypto — and then you should leave. Custody on any venue above means the keys are not yours, and the strongest evidence in this whole table is not a fee or a reserve ratio: it is that two of these six had to reimburse customers after something went wrong, and one still has an open case. If your plan is to hold rather than trade, buy on the cheapest venue that legally serves your country, withdraw to a wallet you control, and stop reading exchange rankings — including this one. If you are swapping between crypto assets rather than touching fiat, a decentralised exchange removes the custodian from the transaction entirely, at the cost of gas and managing your own keys. Neither route earns us anything, which is exactly why you are reading it here.
Frequently asked questions
Which crypto exchange is actually the cheapest?+
At the moment we measured, Binance and OKX were cheapest: about 0.10% effective cost on a $10,000 BTC/USDT market buy — essentially the taker fee, with no measurable slippage. Coinbase was the most expensive at roughly 0.62%, more than six times more for the identical trade. But cheapest is not the same as best: the cheapest venue in our table is the one EU users can no longer use.
Can EU users still use Binance?+
Not normally. Binance failed to obtain a MiCA licence, withdrew its EU application on 24 June 2026, and has restricted services to EU/EEA users since 1 July 2026. Withdrawals remain open. If you are in the EU, the fee advantage is moot — you need a venue that is actually licensed to serve you, such as Kraken, Coinbase, OKX, Bybit or Gate, all of which hold MiCA authorisations.
Is Bybit safe after the $1.5 billion hack?+
The February 2025 theft was the largest in exchange history, and the FBI attributed it to North Korean actors. But the attackers compromised a third-party signing tool’s front-end, not Bybit’s own systems, and — the part that matters — no user lost money. Withdrawals were never halted, the gap was closed within about 72 hours, and independent monthly proof-of-reserves audits have shown over 100% coverage since. By our own disqualifier (funds not returned to users), Bybit passes. Its real weakness is elsewhere: direct fiat rails exist in only two currencies.
Why is your cost figure different from the exchange’s fee page?+
Because a fee page is not the price. What you pay is the taker fee plus the spread plus the slippage your order causes as it eats through the book. We compute that by walking each venue’s live order book with a real order size, so the number reflects execution rather than marketing.
Which exchange has the strongest proof of reserves?+
Kraken and Bybit. Both publish a proof-of-reserves that includes client liabilities and is attested by a named third party — Kraken quarterly, Bybit monthly. Binance, OKX and Gate publish cryptographically sound proofs with no independent attestor, and Binance corrected its own reserve-ratio calculation in January 2026 after it had been overstating coverage. Coinbase publishes no crypto proof-of-reserves at all, relying instead on audited public-company financial statements.
Do you earn money if I sign up through your links?+
No. There is not a single affiliate link on this page — every link is direct. Nobody can buy a place in this ranking, accelerate inclusion or influence a score. That is precisely why this page can tell you that the cheapest move might be to not use an exchange at all.
How this is funded
Nothing on this site is for sale. No affiliate links, no paid placements, no sponsored posts, no guest articles — not in this ranking and not anywhere else on the domain. There is no rate card, because there is nothing to buy. Nobody can purchase a rank, accelerate inclusion or influence a score, and no company in this table has ever paid us anything.
That is not a disclaimer — it is the reason this page can put a $4.3bn guilty plea in the fourth row, rank a venue that is 2.6× more expensive in first place, and end by telling you the smartest move may be to use no exchange at all. Full funding policy · Methodology