We re-checked 12 fee schedules

Every fee on this site carries a date. That date is the only thing separating a comparison table from a guess, because exchanges revise fee schedules without announcement and comparison sites almost never go back to look.

So we went through all twelve venues again against their published schedules. Seven matched what we had. Five did not, and in every case the error ran the same direction: the number in wide circulation was lower than the number a new account actually pays.

That direction is not a coincidence, and it is worth understanding before the specifics.

Most exchanges publish several rates for the same trade. There is the standard tier. There is the rate after you hold the exchange's own token. There is the rate after you buy or issue a fee voucher. There is the VIP ladder, which depends on thirty-day volume you do not have on day one.

The lowest of those numbers is the most attractive to quote, and it propagates. One site writes it down, the next copies it, and within a year it is simply what everyone says the fee is. Nobody is necessarily lying. The number just stopped being labelled somewhere along the way.

The result is a table that ranks venues using rates which different readers would need entirely different conditions to obtain. That is not a comparison at all.

This is the largest error we found, and one we had made ourselves. Gate's spot fee is quoted as 0.09% maker and taker almost universally, and we had recorded exactly that.

The VIP 0 rate. No GT holding, no volume history, what a new account pays on its first trade — is 0.20% for both sides. The 0.09% figure is what you reach after holding GT, Gate's own token, and enabling the GT fee discount.

The size of the error matters more than its existence. At 0.09% Gate would be the cheapest spot venue in this comparison. At 0.20% it is the most expensive, double the 0.10% industry norm. The correction does not nudge Gate's position, it moves it from one end of the table to the other.

Gate's futures pricing is competitive at 0.02% maker and 0.05% taker, matching Binance and OKX. The gap between its spot and futures rates is the widest of any venue here, so whether Gate is cheap depends entirely on which market you trade, and that nuance disappears completely behind a single wrong spot number.

A smaller error, and an interesting one because of how it happened. Bybit charges 0.055% taker on perpetual futures. That half-basis-point increment is unusual, and most fee tables round it to 0.06%.

We had it as 0.06% for a structural reason worth admitting. Our data model stored fees as whole basis points, on the theory that integers avoid floating-point comparison bugs when sorting. The reasoning was wrong: nothing compares these values for equality, because ranking is done on computed annual cost, which is floating point anyway. The constraint bought nothing and quietly rounded a real rate.

Half a basis point sounds like noise, and on one trade it is. It also changes the ordering. At 0.055% Bybit undercuts Bitget and KuCoin, both at 0.06%. At a rounded 0.06% it ties them. A table that rounds cannot show that.

The same failure mode as Gate, in the Korean market. Coinone is routinely described as charging 0.04%, which would put it among the cheapest domestic venues.

The published base rate is 0.20% maker and taker. The 0.04% is the voucher rate. Available to anyone, but only after issuing a voucher, which is a step a new account has not taken.

We record 0.20% for the same reason we record Binance at 0.10% rather than the BNB-discounted 0.075%: the table shows what applies without holding a token or issuing anything. The voucher is real and worth using, and Coinone's page here says so. It is simply not the default.

Korbit is the most interesting case, because the error lies in assuming that one number exists at all.

Korbit lets you choose between two standing fee plans. The Lowest Price plan charges 0.05% on both sides. The Maker Reward plan charges 0.15% taker, and on maker orders charges nothing at all while paying you 0.01% of notional. Neither requires a token or a purchase.

So Korbit's fee is 0.05%, or 0.15%, or negative, depending on a plan you pick and how you trade. Every comparison we checked, ours included, collapsed that into one figure. We had recorded 0.15%, which is the worst of the three for the market-order flow most retail accounts generate.

We now show the Lowest Price plan at 0.05% and explain the Reward plan on the exchange page. That moves Korbit from among the most expensive Korean venues to the second cheapest, behind Bithumb.

The one that surprised us most, because Bithumb is the venue most often called the cheapest in Korea, and 0.04% is stated as its fee almost everywhere.

The published base rate is 0.25%. When Bithumb ended its zero-fee promotion it announced that pricing was being "unified" at 0.04%, and that framing is what propagated. What actually exists is a fee coupon: free to register, applied instantly, valid for thirty days, and then gone unless you register it again.

The renewal is what makes this materially different from Gate's or Coinone's discount. Nothing warns you when the coupon lapses. Your rate silently returns to 0.25%, six times higher, and stays there until you happen to check. A trader buying regularly can pay that for months without noticing.

Correcting it reorders the entire Korean table. On base rates Bithumb is now the most expensive of the four rather than the cheapest, and Korbit's 0.05% Lowest Price plan takes the top spot precisely because it requires nothing to maintain.

Two different answers, depending on how much housekeeping you are willing to do, which is itself the finding.

Do nothing and simply trade: Korbit 0.05%, Coinone 0.20%, Gopax 0.20%, Bithumb 0.25%. Keep every available discount active: Bithumb 0.04%, Coinone 0.04%, Korbit 0.05%, Gopax 0.20%.

Gopax is the venue that should worry you in both columns. It is the only one of the four with no route down from its base rate that we could find, which makes it the most expensive Korean exchange for an attentive user by a wide margin.

This one started as the weakest figure here and ended as the strongest, which is worth recording because of how it was resolved.

Gopax's fee page renders client-side and returns nothing to a fetch, so its 0.20% rested on third-party exchange directories rather than the venue itself. A lower standard of evidence than every other number on this site.

Its public API answers the question directly. `GET /trading-pairs` returns `makerFeePercent` and `takerFeePercent` per pair, and every one of the 115 KRW pairs reports 0.2 on both sides. That is not a screenshot of a marketing page; it is what the matching engine charges, read from the exchange.

The general lesson is worth more than the number. Where an exchange publishes an API, the fee schedule in it is better evidence than the fee page, and it is checkable by anyone reading this.

You do not need to trust this page. The method is short and works on any comparison site, this one included.

These corrections are commercially unhelpful, and it is worth being explicit about that. Gate moving from cheapest to most expensive on spot makes a venue we earn commission from look worse. Correcting Coinone's rate upward does the same.

The alternative is a table that ranks venues by numbers its readers cannot obtain, which is worth nothing to them and, once anyone notices, worth nothing to us either. The dates on every row exist so the check can be repeated, including against us.