Why Bitcoin's Price Differs Across Exchanges, Measured

Bitcoin has no single price because each venue runs its own order book, quotes in its own currency, and moving money between venues costs time and fees. On October 23, 2026, one four-minute reading showed 81,653 on Coinbase, 81,770 on Binance's USDT pair and 114,393 on Kraken's Canadian-dollar pair. After converting the units, the major books sat within about 4.5 basis points (0.045 percent) of each other, with one visible exception explained below.

The three numbers above look 40 percent apart and are not. One is in US dollars, one is in Tether tokens, and one is in Canadian dollars. Most of what readers call a price gap between exchanges is a difference in what the price is measured in. What remains after that is smaller, and it is the part that traders, researchers and index providers argue about.

A dated snapshot: twenty readings of the same coin

The table below comes from public order-book endpoints that need no account. A script requested the best bid and best ask from every venue in the same moment, 20 times, about 12 seconds apart, between 10:25:27 and 10:29:16 UTC on October 23, 2026. Each row shows the median of the 20 mid prices, where the mid price is the average of best bid and best ask. A round of requests took a median of 607 milliseconds and at most 1,002 milliseconds to finish.

Rows are grouped by quote currency and sorted alphabetically within each group. This is a snapshot of a quiet four minutes, not a measure of typical conditions, and the order is not a ranking of anything.

Method notes that matter for reading it:

Excluding Bitfinex's BTC/USD row, the five other dollar books ranged from 81,653 to 81,679 in the median, and the largest gap between the highest and lowest of the five in any single round was 3.5 basis points. Including Bitfinex's dollar book, the same round-by-round gap was 15.3 basis points at the median and 15.9 at the maximum. Bitfinex's BTC/USD row sits about $125 above Coinbase's in the table, which is the one exception the opener mentions. The data point to a reason without proving a cause: Bitfinex's own USDT/USD book (its configuration file labels the code UST as "Tether USDt") showed a bid of 1.0000 and an ask of 1.0001 in all 20 rounds, while Coinbase's and Kraken's books showed USDT trading at 0.9986 to 0.9987 dollars. Converting Bitfinex's BTC/UST mid with Bitfinex's own USDT rate gives a median of 81,769.09 across the rounds, in line with its dollar book. Whether that makes Bitfinex's dollar and USDT books one market is a question about its internals that this snapshot cannot answer.

Units first: why a USDT price is not a dollar price

Tether's token, USDT, is meant to track a dollar, but it trades as its own asset against dollars. On the Coinbase and Kraken books in the snapshot it changed hands at 0.9986 to 0.9987 dollars, a little under par. Many bitcoin order books are quoted in USDT rather than in dollars: seven of the fifteen rows in the table are. A bitcoin priced at 81,776 USDT is priced at 81,776 tokens, and each token is worth whatever the USDT/USD book says it is.

The worked example uses one round from the snapshot, round 10, taken at 10:27:15 UTC on October 23, 2026.

Read naively, Binance's bitcoin is $116.30 dearer than Coinbase's (81,775.995 minus 81,659.695), a gap of 14.2 basis points. Multiply Binance's USDT price by the dollar value of one USDT: 81,775.995 times 0.998635 is $81,664.37. The difference from Coinbase's dollar book falls to $4.68, or 0.57 basis points. About 96 percent of the apparent gap was the unit. Run the other way, the dollar-equivalent of one USDT implied by the two bitcoin books is 81,659.695 divided by 81,775.995, which is 0.99858, close to the 0.99863 shown by Coinbase's own USDT book at the same moment.

The effect is not a one-minute accident. The same comparison was run on one-minute closing prices (the last trade in each minute) for the seven days from October 16 to October 23, 2026, using Coinbase BTC-USD, Bitstamp BTC/USD, Binance BTC/USDT and Coinbase USDT-USD. Of 10,080 minutes, 10,030 had a one-minute candle in all four series and are used here. That is a different basis from the order-book mids above: it uses trade prints, so each series carries some bid-ask noise of its own.

Over those seven days, USDT's close on Coinbase ranged from $0.99849 to $0.99928, and its median was 12.40 basis points below a dollar. The number printed on a USDT-quoted chart therefore ran roughly 13 basis points above the dollar-quoted number for most of the week without bitcoin being any cheaper anywhere. Our guides on how stablecoin types differ and on USDT and USDC market share explain what each token is, and our piece on MiCA and stablecoin liquidity fragmentation covers how rules can split dollar-token liquidity across venues.

This is also the first place where a comparison of prices goes wrong in practice. A table that lists "Binance BTC 81,776" next to "Coinbase BTC 81,660" without saying USDT versus USD makes the first look 0.14 percent dearer. The fix costs one column: always write the quote currency and the time.

Euros, Canadian dollars and what the exchange rate is

Quotes in euros or Canadian dollars have the same problem in a different form. Kraken's XBT/EUR mid in the snapshot was 72,924.90. Multiplied by Kraken's EUR/USD mid of 1.11973, that is 81,658.07 dollars. Kraken's XBT/CAD mid of 114,393.27 divided by its USD/CAD mid of 1.40115 is 81,643.00 dollars. Measured against Coinbase's dollar mid, the euro row is 0.6 basis points higher and the Canadian-dollar row is 1.3 basis points lower. Both gaps are smaller than the 6.1 basis-point market move across the four minutes.

Two cautions apply. The conversion rate has its own source and time. A foreign-exchange rate pulled from a different provider, or a daily reference rate published once an afternoon, would add a mismatch that has nothing to do with bitcoin. And a euro or Canadian-dollar quote on a given exchange comes with that exchange's deposit routes in that currency. A bitcoin you buy with local currency can cost more than the dollar-converted global price when the route into that currency market is narrow, as the capital-control cases in the research below show at extreme scale.

What a reference rate is and how one is built

Because venues disagree slightly, markets that need one number use a reference rate: a rule that collects trades or orders from approved venues and reduces them to a single figure. The CF Benchmarks Bitcoin Reference Rate (BRR) is one widely used example. CF's own page describes it as a once-a-day benchmark that aggregates trade data from Bitcoin-USD markets on exchanges meeting its Constituent Exchange Criteria. It has been calculated every day since November 14, 2016, and is the settlement index for futures listed by CME Group and Crypto Facilities MTF and a pricing source for net asset value determinations of investment products from major financial institutions including WisdomTree Europe. On the day this article was researched, the page showed $81,662.71, last updated Thursday, October 22, 2026, 15:00:00 GMT. That one-per-day rate is a different thing from the live prices in the snapshot above, and the two should not be set side by side as if they described the same minute.

The methodology guide (version 17.4, dated August 24, 2026) gives the recipe for the standard BRR:

The guide's list of constituent exchanges for the Bitcoin rates, in its version 13.7 dated August 24, 2026, names Bitstamp, Coinbase, Kraken, Gemini, Bullish Exchange, Crypto.com and LMAX Digital as listed without a suspension date, and shows itBit as suspended on July 27, 2026. Bitfinex and OKCoin.com (HK) appear as added in November 2016 and suspended in April 2017. CF verifies venues against its published criteria under an oversight committee, and the list changes, so a reference rate is a statement about a chosen set of venues and not about every market.

Why medians, shown with a hypothetical partition

The guide explains the design choices in its own words. Taking a median rather than an average "greatly reduces" the rate's susceptibility to price extremes on one or more constituent exchanges, weighting by volume means that whether an order is executed in parts or in full has no effect on the result, and averaging the partitions equally means a single large trade or cluster of trades in any one partition has only a limited effect.

Here is what that means in numbers. The partition below is hypothetical: seven trades from several venues in one five-minute window, all priced between 81,645 and 81,660, totalling 10.2 BTC. It uses the common definition of a volume-weighted median, the price at which cumulative volume reaches half of the total when trades are sorted by price. CF's published formula (its Equation 1) is the authority on tie handling, so treat this as an illustration of behaviour, not a replication of CF's code.

A half-bitcoin print at a price $248 off the market moves the average by about $11.6 and the median by nothing. Even a 6-bitcoin print, which moves the mean by $91.7, moves this partition's median by $3. And because the BRR then averages 12 partitions with equal weight, an event confined to one partition and moving its median by $3 moves the final rate by $0.25. That is the reason reference rates are built this way: they are meant to be hard to move with one venue's odd print, which also means they will not match any single venue's price at any given second.

CF also publishes a Real Time Index for bitcoin (BRTI), which works from order books instead of trades. Its guide says it consolidates the order books of the constituent exchanges, removes crossed orders, weights the mid-price curve by an exponential distribution up to a computed depth, and sums it. A once-a-day trade-based rate and a continuous order-book-based rate are separate products from the same provider.

Another reference rate, built differently

Reference rates are choices, and different providers choose differently. The snapshot's cross-check series is Coin Metrics' one-second USD reference rate, which is available on a free community API. Its published methodology lets a non-stablecoin asset draw on spot markets quoted not only in US dollars but also in Bitcoin, Ethereum, Tether, USD Coin or Wrapped Ether, converting each trade to dollars using the reference rates of BTC, ETH, USDC and USDT. At real-time frequencies it takes the most recent trade from each constituent market and calculates a weighted median in which half the weight comes from volume over the previous 60 minutes and half from inverse price variance over the same period. Markets whose daily volume-weighted price deviates more than 3 percent from the median are excluded.

Two consequences follow. A stablecoin-quoted market can feed one reference rate and be absent from the other, so even "the reference price" depends on which provider. And the Coin Metrics figure used as a cross-check in the snapshot partly measures agreement with the median of the markets it includes, which already contain USDT pairs. It is a convenient yardstick, not independent proof of what bitcoin "really" cost. CF's page says the BRR is also used for net asset value determinations of investment products. That connects to bitcoin ETF premiums and discounts, where the gap between a fund's market price and its asset value is the same kind of gap, one level up.

Why traders do not close the gaps instantly

If bitcoin were cheaper on one venue by 0.15 percent, a trader could buy there and sell on the other. In textbooks that trade pushes the two prices together. In practice several frictions stand in the way, and the published research has measured them.

The New York Fed's Liberty Street Economics post by Alexander Kroeger and Asani Sarkar (March 23, 2016) looked at bitcoin-dollar prices on BTC-E, Bitfinex and Bitstamp and found that bitcoins on BTC-E "consistently trade at a discount" to the other two, a discount that "averages about 2 percent and has at times been higher than 20 percent." Their explanation is a list of costs. BTC-E charged 0.2 to 0.5 percent per transaction, with additional fees to deposit or withdraw traditional currency, then $20 for a wire deposit. Wire deposits took five to ten days, and moving bitcoin onto the other exchanges needed three network confirmations, about thirty minutes. The post notes that the intraday volatility of the BTC-E price often exceeded the average price difference, so waiting for transfers could erase the gap before it was captured. It adds counterparty risk: eighteen of forty bitcoin exchanges in a 2013 study had ultimately failed, and Mt. Gox lost roughly $460 million of customers' bitcoin in 2014. The bid-ask spread, by contrast, was "negligible relative to the typical price difference."

Those figures are from 2016 and describe one venue's circumstances. They are cited for the structure of the argument, not as current withdrawal times.

Makarov and Schoar, in "Trading and Arbitrage in Cryptocurrency Markets" (Journal of Financial Economics, 2020, using the accepted manuscript at the LSE repository, which may differ from the typeset version), describe the same constraint in their own terms. A textbook arbitrage would be to buy bitcoin in the US, sell it for Korean won in Korea, convert the won to dollars and send the dollars home. They write that "In practice, this textbook arbitrage is not possible", because a bitcoin transaction takes about an hour to register on the blockchain and exchanges "typically take from a few hours to several days to transfer fiat currency." To lock in a profit the trader has to buy on one venue and sell on the other at the same time, holding inventory of both bitcoin and cash in both places, which brings back the risk that the prices diverge further before converging. They conclude that, because of capital controls, "arbitrageurs find it difficult to scale up their trading."

A hypothetical trade, priced

The snapshot gives a way to see what a simultaneous trade would need to clear. The median Coinbase ask in the 20 rounds was $81,653.25 and the median Bitfinex bid was $81,771.50. Buying at one and selling at the other would have earned a gross $118.25 per bitcoin, or 14.48 basis points of the buy price. The opposite direction, buying at Bitfinex's ask and selling at Coinbase's bid, would have lost $133.26.

Now add fees. This is hypothetical and uses Kraken's published spot taker schedule, read on October 23, 2026, only to show scale. These are not the fees of Coinbase or Bitfinex and say nothing about what those firms charge.

At the entry tier, the gap is smaller than the fee by a factor of eleven. At the top tier, a positive margin remains only for a firm with large capital on both venues, which still faces the transfer and balance constraints that Makarov and Schoar describe and the counterparty risk the New York Fed post describes. During the 3.8 minutes of the snapshot, the market itself moved 6.1 basis points on the Coin Metrics series, more than two-fifths of the 14.5 basis-point gap. Our exchange fees guide covers the commission layers a retail trader faces, and the slippage guide covers the cost of moving the order through the book, which a one-bitcoin hypothetical ignores.

Rules also split markets

The Federal Reserve Bank of Dallas working paper by Gina Pieters and Sofia Vivanco (No. 293, December 2016) gives a different reason. The abstract reports "systematic differences in bitcoin prices across 11 different markets representing 26% of global bitcoin trade volume" and concludes that markets that "do not require customer identification for establishing an account are more likely to deviate from representative market prices than those which do." Its sample is daily data from June 1, 2014 to July 31, 2015 for exchanges quoting in US dollars. The authors read this as evidence that know-your-customer rules affect pricing, which is a claim about a 2014 to 2015 sample and not about today's venues. For how those rules look now, see our explanation of the Travel Rule.

The price in your app is a different object

Apps add another layer. The number on an app's chart and the number you pay can differ, and neither needs to match a venue's order book.

Robinhood's support page on crypto order routing says that orders "are filled at the bid or ask price, not at the mid price," and that the mid price "is the price shown on the chart." The page gives a $100 example with a 0.96 percent buy spread, of which Robinhood receives $0.95, effective June 15, 2026. Kraken's fee page says that on its Instant Buy and Sell services "the price you see includes a spread, the difference between the market rate and the rate you receive," and that the spread "may vary for similar transactions," depending on volatility, asset, order size and other factors.

Put the two together and an app's buy quote can sit well away from a venue's mid without the asset being priced differently. With the Coinbase mid of $81,653.25 from the snapshot, a 0.96 percent spread on the buy side, the page's own example figure applied as a hypothetical, would put a buy fill at $82,437, which is $784 above the venue's mid. That is a service charge for immediate execution at a quoted price, not a second price for bitcoin. The fees guide explains how to measure your own app's spread against an order book in two minutes, and the slippage guide explains why large orders on a venue's book also fill away from the mid.

What the research found when gaps were enormous

The following findings are from published work, each with its own period and basis. They should not be compared with each other or with the snapshot, because each measures something different.

Makarov and Schoar's measurement is worth understanding, because the same formula is easy to apply to any set of prices. Their "arbitrage index" is the maximum price across exchanges divided by the minimum, computed per minute from volume-weighted average trade prices and then averaged by day. An index of 1.05 means the highest venue was 5 percent above the lowest. For the four main US exchanges they studied (Bitstamp, Coinbase, Gemini and Kraken), the index was low most of the year, with average price dispersion below 2 percent, and around 1.04 in a few weeks of March, June and December 2017. Within Europe the index averaged about 1.02. Across regions it was far larger: Japan against the US about 10 percent, US against Europe about 3 percent, US against Korea the 15-percent-plus average above. Their text also says that the price gap in the dollar-to-bitcoin quote between the US and Korea was above 20 percent while the ethereum-to-bitcoin price difference was 3 percent on average over the same period, a contrast they cite in support of the importance of capital controls.

Our own per-round max-over-min in the snapshot is the same formula on a different basis, applied to quoted mids across six dollar books in four minutes, and gives a median of about 1.0003 without Bitfinex's dollar book and about 1.0015 with it. It would not be right to call that smaller or larger than a historical index, because the inputs, venues and period differ, and the paper links wider deviations to times of large bitcoin appreciation. Makarov and Schoar also found that opportunities were not permanent: in their data an arbitrage opportunity in any given market was open for less than two days on average and never more than a month.

The one steady gap in our own data was USDT sitting a median 12.4 basis points under a dollar for a week on Coinbase. The stablecoin redemption guide is the next read for how dollar tokens are redeemed and why that matters for their price, and the market structure hub collects our other guides on how venues price and fill orders.

Frequently Asked Questions

Why is the Bitcoin price different on Coinbase and Binance?

Usually because they quote in different units. The Coinbase book in our snapshot was bitcoin against US dollars, while Binance's was bitcoin against USDT. In our October 23, 2026 snapshot, Binance's USDT price ran about 14 basis points above Coinbase's dollar price, and multiplying by the USDT/USD rate removed all but about 0.6 basis points of it. Small residual differences come from separate order books and timing.

Which exchange has the real bitcoin price?

None of them is the real price, and the question assumes a single price that does not exist. Each venue's order book is a real, tradable price for its own pair at its own moment. If you need a single figure, a reference rate such as CF Benchmarks' BRR or Coin Metrics' real-time rate collects trades from several venues and applies a published rule, and the two rules differ.

Can you make money by buying bitcoin on one exchange and selling on another?

Rarely at retail scale. The gap has to exceed the trading fees on both sides, the cost of moving cash or bitcoin, and the risk that prices shift before the trade completes. In the snapshot, a 14.5 basis-point gap would have been wiped out by two taker fees at an entry-level schedule. Researchers found that capital controls and transfer delays kept some gaps open for days. Nothing here is a trading suggestion.

Why does my app show a higher price than the exchange chart?

Most likely because the app's quote includes a spread. Robinhood's support page says orders are filled at the bid or ask rather than the mid price shown on the chart, and Kraken says the price on its Instant Buy and Sell services includes a spread. The difference is a service charge, not a different price for bitcoin. Our fees guide shows how to measure it.

What is USDT, and why does a USDT price differ from a dollar price?

USDT is a stablecoin intended to track the US dollar, but it trades as its own asset. In the snapshot, Coinbase's USDT-USD book showed about 0.9986 dollars per token, while Bitfinex's own book showed 1.0000. A bitcoin priced in USDT is priced in those tokens, so it has to be converted at the USDT/USD rate before it can be compared with a dollar price.

What is the CF Benchmarks Bitcoin Reference Rate?

It is a once-a-day bitcoin price published by CF Benchmarks. Per the methodology guide, it takes trades from approved exchanges in the bitcoin-dollar pair during the hour before 4:00 p.m. London time, splits them into twelve 5-minute partitions, takes a volume-weighted median in each, and averages the twelve. CME Group and Crypto Facilities MTF futures settle against it.

Did bitcoin prices differ more in the past?

Yes, in documented cases. The New York Fed's 2016 post found BTC-E averaging about 2 percent below Bitfinex and Bitstamp and at times more than 20 percent. Makarov and Schoar found the US-Korea daily average ratio above 15 percent from December 2017 to early February 2018, reaching 40 percent for several days. Those are findings from specific periods and venues, not a forecast.

Sources

CryptosEyes publishes general educational research, not investment, legal, or tax advice. The snapshot and worked examples are illustrations of how prices are quoted and are not a recommendation about any venue, token or trade, and none of it predicts a price.