# Best Crypto Exchange for Algorithmic Trading in 2026 Updated: 24 September 2026 / Tested: 22 September 2026 edgeX held the deepest resting book in the window we polled to 22 September 2026, a median $29.3 million within 10 basis points of mid on the BTC perpetual, so it leads this page. The best crypto exchange for algorithmic trading is the one whose book absorbs the size you trade; leverage magnifies every mistake a strategy repeats. Weights: What the venue automates for you 24 · Programmatic access 14 · Measured liquidity 22 · Round-trip cost 16 · Security and record 12 · Custody and settlement 8 · Market coverage 3 · Access 1. One formula, fixed 23 September 2026: https://ellsworthvane.com/rating-method ## Best crypto exchange for algorithmic trading, ordered by measured book depth The order is the median depth resting within 10 basis points of mid on each venue's BTC perpetual, from our own snapshots. The mark out of 40 is the site formula and rides along for reference; it does not set the sequence here, and the highest of those marks is Aster's 30.0, in place 4. | # | Venue | Median depth, 10 bps | Score of 40 | Maker / taker | Perp markets | Model | |---|---|---|---|---|---|---| | 1 | edgeX | $29.3 million | 24.3 | 0.040% / 0.045% | 172 | Hybrid | | 2 | EVEDEX | $19.0 million | 29.1 | 0.015% / 0.045% | 52 | Hybrid | | 3 | Lighter | $17.4 million | 28.6 | 0% / 0% | 214 | Hybrid, ZK-proven | | 4 | Aster | $14.6 million | 30.0 | 0% / 0.040% | 578 | Own layer 1, operator-run | | 5 | Hyperliquid | $8.1 million | 24.8 | 0.015% / 0.045% | 324 | On-chain order book | | 6 | Extended | $7.7 million | 27.5 | 0% / 0.025% | 325 | Hybrid | | 7 | ApeX Omni | $1.7 million | 27.0 | 0.020% / 0.050% | 125 | Hybrid | | 8 | dYdX | $848,659 | 22.3 | 0.010% / 0.050% | 78 | App-chain, validator-held book | Facts as of 18 September 2026; entry-tier fees on the BTC perpetual. Access differs by country. ## Read-offs 1. Order by resting depth and the board turns over: Aster tops the site-wide mark at 30.0 and sits fourth here, behind edgeX, EVEDEX and Lighter. 2. The same $100,000 order cost 0.02 bps where the book was deepest and 6.79 bps where it was thinnest. 3. Sub-accounts do work that margin modes do not: edgeX documents up to 20 per wallet, Extended up to 10 and no hedge mode. ## 1. [edgeX](https://pro.edgex.exchange) — best for the deepest resting book Score 24.3. edgeX held a median $29.3 million of BTC depth inside 10 basis points of mid, where a $100,000 market order would have moved the price by 0.02 bps. Up to 20 trading accounts sit under one wallet, each with its own margin, which is how a long and a short in one market are held apart ([edgeX documentation](https://edgex-1.gitbook.io/edgeX-documentation/trading/trading-accounts-and-margin), checked 18 September 2026). - Works: Median BTC depth of $29.3 million inside 10 basis points; Up to 20 trading accounts per wallet, each with its own margin. - Falls short: No rate limit published anywhere: the page describes the mechanism and prints no number; The maker side is 0.040%, above every other venue here. - Not for: a strategy that pays the maker side all day. ## 2. EVEDEX — for off-chain matching, periodic on-chain settlement Score 29.1. EVEDEX is a hybrid perpetual futures exchange: matching off-chain in a central order book, settlement on-chain on Arbitrum, which takes a position snapshot once enough has changed (the EVEDEX documentation we read on 18 September 2026). Its BTC book held a median $19.0 million inside 10 basis points of mid, second here. Its developer documentation publishes a rate limit as a number — 30 heavy requests per 60 seconds per account, alongside up to 500 active orders and 500 active take-profit and stop-loss orders — and four strategies of its own run on the BTC, ETH, SOL and XRP perpetuals. - Works: Median BTC depth of $19.0 million inside 10 basis points, on 117 snapshots; A rate limit published as a number, and four strategies the venue wrote itself. - Falls short: Reduce-only is documented nowhere, and post-only appears in the interface specification as a bare field with no values, no default and nothing said about an order that would cross; Positions reach Arbitrum in periodic batches rather than on every fill; Funded subaccounts are browser-only, so a bot cannot be run on one. - Not for: a strategy that must post without ever crossing. ## 3. [Lighter](https://lighter.xyz) — best for a zero-fee tier with a latency price Score 28.6. Lighter's Standard account pays 0% maker / 0% taker and delays taker orders by 300 ms; the opt-in Premium tier charges a fee and cuts that to 140 ms ([Lighter documentation](https://docs.lighter.xyz/trading/trading-fees.md), checked 18 September 2026). The documentation also describes a priority queue on Ethereum through which traders exit with proofs if the sequencer stops answering them. - Works: No maker or taker fee on the default Standard account; A documented withdrawal route that works without the sequencer. - Falls short: 300 ms added to every taker order at the zero-fee tier; Every order passes through one operator-run sequencer. - Not for: strategies that cannot pay for the faster tier. ## 4. [Aster](https://www.asterdex.com) — best for the shortest block interval Score 30.0. Aster matches and settles on its own proof-of-stake chain, live since 16 March 2026, and its documentation puts block latency at 50 ms ([Aster documentation](https://docs.asterdex.com/overview/what-is-aster.md), checked 18 September 2026). For a strategy that cancels and replaces constantly, that interval decides whether an order rests or is already stale. Its Pro book also held the narrowest median spread of the eight at 0.01 bps, a figure Lighter matched. - Works: Documented block latency of 50 ms, with post-only, reduce-only and a published rate limit; 578 perpetual markets, the widest count here. - Falls short: Core chain code is closed-source, with no external validators in phase one; Published audits cover vault and earn contracts, not the chain. - Not for: traders who want the matching layer open to inspection. ## 5. [Hyperliquid](https://hyperliquid.xyz) — best for size, on a public chain Score 24.8. Hyperliquid commits every order, cancel and liquidation to its own chain ([Hyperliquid documentation](https://hyperliquid.gitbook.io/hyperliquid-docs/about-hyperliquid.md), checked 18 September 2026). It carried $8.7 billion of open interest on $13.4 billion of 24-hour volume on CoinMarketCap on 22 September 2026, the largest figures here and a fair illustration of why they are not the ranking column. - Works: Every order, cancel and liquidation committed to the venue's own chain; Bug bounty of up to 1 million USDC. - Falls short: Its own vault absorbed about $4.9 million in the November 2025 POPCAT episode; 146 of 324 markets are deployed by third parties. - Not for: strategies that assume every ticker on the screen was vetted. ## 6. [Extended](https://extended.exchange/) — best for a wide list, quoted on request Score 27.5. Extended settles every transaction on Starknet and charges 0% maker / 0.025% taker on every account ([Extended documentation](https://docs.extended.exchange/extended-resources/trading/trading-fees-and-rebates), checked 18 September 2026). Its BTC book held $7.7 million, below every venue above it here, the public book covers 47 of 325 active perpetuals, and there is no hedge mode at all. - Works: Bug bounty of up to $500,000 for critical reports; 0% maker fee with no volume tier to reach first. - Falls short: The public book covers 47 of 325 markets; the rest are quoted on request; No hedge mode, so a long and a short need separate sub-accounts. - Not for: strategies that hedge inside a single market. ## 7. [ApeX Omni](https://omni.apex.exchange/) — best for a thin book with a way out Score 27.0. Matches on ApeX Omni are proven to zkLink contracts on the rollup, and the documentation publishes a force-withdrawal route for the day the operator stops answering ([ApeX Omni documentation](https://docs.zk.link/architecture/transactionflow), checked 18 September 2026). The book is the weakness: a median $1.7 million inside 10 basis points of mid, against $29.3 million at the top of the page. - Works: A force-withdrawal route to layer 1 is documented; A $100,000 order cleared at 0.39 bps despite the thin book. - Falls short: 0.050% taker fee, matched only by dYdX for cost here; Median BTC depth of $1.7 million, below every venue above it. - Not for: strategies that move size at short notice. ## 8. [dYdX](https://www.dydx.xyz) — best for a book held in validator memory Score 22.3. dYdX v4 keeps orders in validator memory and commits matches on its own chain ([dYdX documentation](https://docs.dydx.xyz/concepts/architecture/overview), checked 18 September 2026). Its BTC book was the thinnest of the eight, and a $100,000 market order would have moved the price by 6.79 bps, more than three hundred times the cost of that order against the deepest book here. - Works: A bug bounty of up to $1,000,000 through Cantina since 8 May 2026; Orders live in validator memory, not in one operator's engine. - Falls short: The thinnest book here, a median $848,659 inside 10 basis points; Transfers and withdrawals gate for 50 blocks after a five-minute outage. - Not for: strategies sized above a few tens of thousands of dollars. ## Why the order here is depth and not volume A strategy that rests orders does not consume a headline. It consumes whatever sits near the mid price at the millisecond it fires, and that is what this page is sorted on. Across the eight the median depth inside 10 basis points of mid ran from $29.3 million down to $848,659, and one $100,000 market order would have moved the price by 0.02 bps at one end and 6.79 bps at the other, over the same window. Self-reported turnover is the figure we refuse to rank by. Hyperliquid published $13.4 billion of 24-hour volume on CoinMarketCap on 22 September 2026 and Aster $3.2 billion, about a quarter as much, yet the two BTC books we read held $8.1 million and $14.6 million. ## What the formula cannot score Eight criteria produce the mark out of 40 — two of them added on 23 September 2026, for what a venue automates and what its order interface documents — and an unattended strategy still meets a layer underneath made of conditions rather than degrees. Whether a venue runs a hedge mode decides how many accounts exist before the first order goes out. How fast its blocks close decides whether a cancel is effective or merely accepted. What it does after an outage decides what a strategy finds when it comes back. Those answers sit on the cards above. ## Where every figure on this panel came from Eight criteria, fixed on 23 September 2026 before any venue here was marked under them, applied to 8 venues ([how we rate](/rating-method)). The sequence is not the mark: it is the median depth resting inside 10 basis points of mid on each BTC perpetual, polled at ten-minute spacing, the last reading dated 22 September 2026. Fees, order interfaces and account limits come from each venue's own documentation. The snapshots cover one instrument, the BTC perpetual: a venue deep in BTC can be thin in the market a strategy actually trades, and a single day's loop of the books is not an hour of news. Open interest and 24-hour volume are each venue's own figures, republished by CoinMarketCap on 22 September 2026. ## FAQ ### Which crypto exchange is best for algorithmic trading? On our readings to 22 September 2026, edgeX: a median $29.3 million of BTC depth resting inside 10 basis points of mid, the deepest of the eight. Depth is not the whole answer — a strategy hedging inside one market needs sub-accounts, and Extended runs no hedge mode. ### How much order book depth does an algorithmic strategy need? Enough that the largest single order is a small fraction of it. A $100,000 market order would have moved the price by 0.02 bps against the deepest book here and 6.79 bps against the thinnest, in the same window and the same market. ### Is trading volume a good measure of liquidity? Not on its own, and every venue reports its own. Hyperliquid published $13.4 billion of 24-hour volume on CoinMarketCap on 22 September 2026 against Aster's $3.2 billion, yet the two BTC books we measured held $8.1 million and $14.6 million inside 10 basis points of mid. ### Can I run a long and a short in the same market at the same time? Only where the venue runs a hedge mode or lets positions be split across sub-accounts. Extended states it has no hedge mode, so the two legs need separate accounts and a wallet gets up to ten. edgeX documents up to 20 accounts per wallet, each with its own margin. ### Do crypto exchanges allow trading bots? All eight venues here publish market data anyone can poll, and none forbids automation in the terms we read on 18 September 2026. What varies is what an automated account can reach: a funded subaccount on EVEDEX runs in a browser only, so a bot cannot be pointed at one. ### How do perpetual venues handle a chain outage? Badly, from a strategy's point of view. dYdX gates subaccount transfers and withdrawals for 50 blocks after an outage of five minutes or longer, and Extended restarts into 120 seconds in which market orders and stop triggers do not work. Positions stay open throughout. ### Does the maker or the taker fee matter more for an algorithm? It depends on whether the strategy rests or crosses. A resting strategy lives on the maker side, where Aster, Extended and Lighter charge nothing and edgeX charges 0.040%. A crossing strategy pays the taker side twice a cycle, from $0.00 to $10.00 on $10,000. ### What is price impact on a perpetual futures order? The distance between the average price a market order actually fills at and the mid price before it was sent. We work it out from each venue's resting book at a fixed order size, so it is arithmetic on published orders rather than a claim about execution. ### How many sub-accounts can one wallet hold? It varies by venue, and the documentation says so plainly. edgeX allows up to 20 trading accounts under one wallet, each with its own margin; Extended allows up to ten. dYdX separates them differently: an isolated-market position locks that subaccount until the position is closed. ### Why is the ranking on this page not the site score? Because the page asks one question and the formula asks eight. The sequence here is the measured median depth inside 10 basis points of mid, and nothing else. The mark out of 40 keeps its own column so a reader can see where the two disagree, which on this page is often. ## Quote bank > "There is no public order book. The order book displayed on the trading screen is indicative only and does not represent resting orders from other traders." — Extended documentation, Request-for-quote execution, 18 September 2026. https://docs.extended.exchange/extended-resources/trading/rfq-execution > "Currently, our platform does not support hedge mode, which would allow users to simultaneously open long and short positions in the same market." — Extended documentation, Trading accounts and modes, 18 September 2026. https://docs.extended.exchange/extended-resources/trading/trading-accounts-and-modes > "All subaccount transfers and withdrawals will also be gated for 50 blocks if a 5+ minute chain outage occurs." — dYdX documentation, Withdrawal limits, 18 September 2026. https://docs.dydx.xyz/concepts/trading/limits/withdrawal-limits Corrections: editorial@ellsworthvane.com. Signal Grid research, 24 September 2026 These panels are paid placement.