The same coin is simultaneously valued differently across exchanges. Arbitrage turns this gap into profit: buy where it's cheaper, sell where it's more expensive. However, discrepancies rarely exceed fractions of a percent and last only seconds. By the time a trader notices the spread and manually places two orders, the gap has already closed.
The inter-exchange arbitrage platform Arbitron takes over this work. The service monitors order books and funding rates on 20 exchanges, and places both sides of the trade—what traders slangily call legs—itself. Funds remain on the trader's exchange accounts.
Money on price differences
Each exchange has its own order book, its own buyers and sellers, its own regional demand. Liquidity flows between venues unevenly, and deposit and withdrawal fees widen the divergence in quotes.
An arbitrageur doesn't guess where the price will go. Two opposing positions cancel out price risk: no matter which way the market moves, one leg compensates for the other. Earnings come from the spread and the funding rate. On paper it all looks simple, but in practice it comes down to speed and coverage.
"A typical inter-exchange spread is 0.01–0.5% and lasts seconds. To consistently earn on such gaps, you need to monitor thousands of pairs across dozens of venues and send two orders simultaneously. A human physically can't keep up: while they're looking at the number, the order book has already changed," — explain at Arbitron.
Four gaps between the screen and the account
The number the scanner draws is raw: it's just the difference between two quotes. What remains of it before real profit is noticeably less. Four factors eat away at the spread: taker fees on both legs, order book depth for the required volume, slippage during order execution, and the decay of the spread itself.
Fees are the easiest to calculate: a full two-legged arbitrage cycle is four market orders (open both legs and close them). Venue tariffs vary significantly, so choosing the right pair matters a lot. Meanwhile, the spread lives exactly as long as it takes other arbitrageurs to close it. A scanner updating every 10–15 seconds already shows history, and the slowest link in the chain remains the human.
23 milliseconds to Binance
Arbitron is connected to every trading pair on 20 exchanges, which means about 10,000 order books simultaneously. The platform holds the entire order book in memory, not just the top quote. At the best price, there's often only a couple of dollars of volume, and on a real order the spread disappears. The service calculates the price at which an order will actually execute for $25, $100, $500, or $1000.
Each account gets a dedicated AWS server with its own static IP, and the trader chooses the region themselves: Tokyo, Singapore, Frankfurt, or London. One relocation per month is free. Most major venues keep their engines in several Asian cloud regions. Binance, Gate, KuCoin, Bitget, HTX, and CoinEx respond fastest from AWS Tokyo, where MEXC and Hyperliquid also operate. Bybit and Phemex are hosted in Singapore. OKX is closest to Hong Kong. Derivatives veterans live in Europe: Deribit's engine is in London, and Poloniex also responds fastest from Europe.
From a region neighboring the exchange, a request takes 10–35 ms, while from another continent it's 200–500 ms. In that time, the order book of a liquid futures contract manages to fully update, and the order arrives at a book that no longer exists. In the team's measurements, such a trade cost nearly 0.8% in slippage—several times more than a typical spread.
Exchanges also limit the number of requests from a single address, and when working from a shared pool, all service clients burn through the limit at once. Others' activity directly slows execution, while a personal IP removes that dependency. The trading core is written in Rust: according to the developers, the language was chosen for its lack of sudden pauses on the critical path. The interface is modeled after an exchange terminal. The price is highlighted with every change, and if the data stream breaks, the number dims and is struck through.
Backtest instead of a showcase
The platform has two search tools. Signals show real-time data: as soon as the gap crosses a set threshold, the trader gets a notification.
The scanner works differently from a typical screener: the table shows not the current spread, but the result of a backtest with a 5-minute delay. The platform runs the last 8 hours of recorded quotes through the same two-threshold strategy that trading cards use. Each row is the result of such a simulation, with fees already deducted.
The scanner calculates the taker fee not by the market average, but by the personal tariff: the trader specifies their VIP level and native token discount in the exchange key settings. The breakeven threshold is measured from that number. The backtest also factors in delay: the trader selects an interval of 0.5, 1, or 2 seconds, and each signal is calculated at the worst price over that period. A spread that flashed for half a second doesn't make it into the report. Along with it, the main illusion of arbitrage screeners is filtered out: pretty percentages that are out of reach.
Next to each opportunity is a reliability score from 0 to 100: the number of cycles closed by the backtest, the profit margin relative to fees, and the daily volume of the less liquid leg. A third of the score comes from market depth, so a thin pair has a hard time getting a high rating. Anything below 50 is suggested in the knowledge base to be treated as a yellow flag.
Under each pair lies a 30-day spread history. On the month map, moments when the gap exceeded the chosen threshold are highlighted, and adjacent buttons rewind to the next spike. This shows whether the pair worked all month or just hit a lucky day once. The table also shows when the next backtest result will appear.
Trading cards
Work in Arbitron revolves around trading cards. Each one links a pair of exchanges, a specific coin, and two spread thresholds: for opening and closing. They operate independently, so a failure in one doesn't bring down the others.
The platform calculates the exit threshold together with the cost of reverse trades: the price of reversing both positions is already baked into the profit estimate. Without such an adjustment, the spread looks profitable on paper but actually brings a loss in practice. The card works the divergence in either direction: if the spread goes up, the platform opens in one configuration; if it goes down, in the mirrored one. According to the developers, many bots are rigidly tied to a single scheme (for example, always short on Binance, always long on OKX) and miss half the moves.
Both orders are sent to the exchanges simultaneously, but sometimes one of them is only partially filled. In that case, the algorithm either tops up the missing volume or closes the already opened position. The trader receives a notification about the situation. Before the trade, the card checks whether the order book has enough depth for the entire order: a spread that's real for $500 might turn out to be fiction for $10,000. The "Depth Multiplier" setting pre-scans the book for the required volume and calculates the volume-weighted execution price.
The second filter, the validation window, requires 1000 ms by default: the spread must stay beyond the threshold for a full second without a single dip, otherwise the countdown starts over. Instant threshold crossings don't reach the trade. The trader configures the scanner filter by spreads, venues, and minimum liquidity, or saves a ready-made preset, and then the platform creates and launches the cards itself. Their number is limited both for each set of conditions and for the account as a whole: the automation doesn't exceed the set risk limits. Those who prefer to catch pairs manually are left with a manual mode showing live order books from both exchanges.
Besides the inter-exchange spread, the service calculates funding arbitrage on a spot plus perpetual futures pair, or between two derivatives. There are also inverse contracts, where margin and settlement go in the coin itself rather than USDT: the team notes that few competitors track such pairs.
Position protection: from stop to delisting
The spread doesn't always converge back, so the card has three safeguards. The first is a hard stop: at a set loss threshold, the algorithm exits at market without waiting for a recovery. The second is softer: the platform stops opening new cycles and looks for a convenient moment to close. The third triggers by default at 3% before the liquidation price.
Delisting is more dangerous than any stop: when an exchange removes a contract from trading, one leg disappears while the other remains in the market without insurance. Arbitron catches such events in three layers. The first polls the official announcement APIs of Binance, Bybit, Bitget, and OKX. The second checks the full instrument list every 5 minutes, and the third reacts to a symbol disappearing from the feed. Scheduled exchange warnings are published 3–14 days before removal.
Upon noticing a delisting, the platform halts position building and sends the trader a Telegram notification. Already opened legs can be closed from the regular interface—no need to go to the exchange for that. The service doesn't do this automatically: an exchange API glitch could easily be mistaken for a contract removal and trigger a sale at a random price. Market makers leave a doomed instrument in advance, and the spread widens five to tenfold in the final hours, while funding accelerates to ±1–2% per eight-hour interval. At Arbitron, they advise closing 24–48 hours before the shutdown, while there's still someone in the book to sell to.
Keys that can't withdraw money
The non-custodial model rests not on trust, but on access rights. Arbitron receives an API key with permission only for trading and doesn't ask for withdrawal rights, so it can't take assets.
Arbitron stores keys using an "envelope" scheme with AWS KMS, Amazon's hardware storage. The platform encrypts each record with the AES-256-GCM algorithm. Each user has their own encryption key—there's no single master password for everyone. For each third-party service, a separate key should be issued, withdrawal rights disabled, and it should be tied to a static IP via the exchange's whitelist.
The platform logs all trades in full: price, volume, fee, order number, time. Both the execution speed for each venue and how much one leg lagged behind the other are visible. Balances with open positions are gathered in the portfolio section, and the terminal itself opens in Telegram, with automatic login and a full set of settings.
Pay when you've earned
There are three plans, and they differ not by trading volume, but by what the platform is allowed to do. The Scanner plan at $39 per month opens up data without execution: all scanners at full depth and without delay, the signal feed, live spread stream, depth history, and executable volume. The trader places orders themselves, and there's no profit commission on this plan.
Trader at $99 per month adds what the platform is usually connected for: a dedicated server, a static IP with one free change per month, trading with your own keys, and up to 10 strategies simultaneously. The commission here is 35% of realized profit.
The third plan, Prime at $299, is marked on the pricing page as upcoming. Its commission is lower, 30%, the strategy limit is raised to 50, and the features add one-legged directional trading, multiple accounts on one exchange, and a private Telegram channel. Annual payment for any plan gives two months free: $390, $990, and $2990 respectively.
The platform charges nothing else: not per trade, not for data, not on account capital. Exchange trading fees are paid by the trader to the venues themselves, and they're deducted before the profit commission is calculated. Three additional services are paid separately: VPN, an out-of-turn IP change, and a second relocation between regions within one month.
The platform calculates the commission once a week and only on the realized result: a week without profit costs nothing. A loss doesn't burn away with the reporting period but carries forward. Income that merely covers the hole costs zero, and charges resume only past the previous peak. Minus $100 in the first week and plus $100 in the second—there's nothing to deduct.
Cashback on trading fees applies to seven of the 20 exchanges: 60% on Toobit, 40% on Bitget, BloFin, Gate, and MEXC, 30% on Bybit and KuCoin. However, for this, the account must be registered via Arbitron's link and the UID linked.
Who Arbitron suits
Arbitron is designed for those who understand the basic principles of futures trading and are willing to figure out trading strategy settings. You'll need stablecoins for both legs and accounts on at least two exchanges. Platform representatives recommend starting from $3000, splitting the amount roughly evenly. Initial setup takes from 30 minutes to several hours depending on the number of venues connected. Most of the time goes into creating keys and