Solana sandwich bots took $29.7M from memecoin buyers

01 · A worse priceSomeone bought just before you

You buy a memecoin and get fewer coins than the app showed. Most people blame a busy market. Often a bot got there first.

In mid-September a wallet spent about $240 in SOL, Solana's own coin, on a memecoin. A fraction of a second earlier, another wallet had bought the same coin on the same exchange. About two seconds later, that wallet sold everything it had bought. Its purchase had pushed the price up, so the first wallet got 6% fewer coins than it would have got without it. The other wallet made about $11 on the round trip and kept about $8 after fees.

That pattern has a name: a sandwich attack. The bot's buy and its sell are the two slices of bread, and the other trade is the filling. Bitquery records swaps on most Solana exchanges, including all the big memecoin venues, so we went through 11 months of them, to 7 October 2026, and counted the sandwiches.

02 · The trickWhat is a sandwich attack?

When you swap on a Solana exchange, you trade against a pool: a pot of two coins, say SOL and a memecoin, whose price moves with every trade. Buying the memecoin pushes its price up for whoever buys next. Your app shows a quote and asks how far the price may move before the trade should fail. That setting is called slippage, or slippage tolerance. At 20%, you accept up to 20% fewer coins than the quote.

A sandwich bot lives in that gap. If it learns about your order before it lands, it buys first and pushes the price up. Your trade then fills at a higher price, still inside your limit. The bot sells straight after, at the higher price your own buy created. The gap between its buy and its sell is its profit, and it comes out of your trade.

Solana has no public waiting room for transactions, the mempool that Ethereum has, so a bot has to learn about your trade some other way. A study published in September 2026 by researchers at ETH Zurich, the University of Lisbon, Flashbots and Category Labs found that most Solana sandwiches no longer sit right around the victim. The bot's buy and sell land a few blocks apart, often in blocks made by different validators, the computers that take turns building Solana's blocks. We use that study's rules, plus one of our own, so that our count can be set next to theirs.

03 · The scaleHow many sandwich attacks happen on Solana?

By those rules, sandwich bots carried out 6.9 million attacks on Solana in the 11 months. They caught 2.2 million different wallets between their buys and sells, many of them more than once.

Not all of those wallets belonged to people. About 100,000 of them were bots by Bitquery's other Solana counts: snipers that buy new coins in their first seconds, wallets that swap 10,000 times per month or trade around the clock, and other sandwich bots. Those bot wallets made 37% of the trades that got caught.

The count grew through the year. In November and December last year the bots made about 3,400 sandwiches per day. By September they made 15 times as many, and the first week of October was busier still.

Trading on Pump.fun, the site where most new Solana memecoins start, grew too, but less: Pump.fun swaps per day tripled over the same months, so the attack rate per swap was about 6 times as high. Counted the way the September study counts, which includes the launch trades we leave out, our daily figure for November to June matches the study's own.

04 · The venuesWhere do the attacks happen?

Almost all of them hit memecoins launched on Pump.fun. 96% of the sandwiches landed in a coin still on Pump.fun's launch curve, the pricing formula a new coin trades on at first, or in a coin that had moved on to PumpSwap, Pump.fun's own exchange.

Memecoins suit the bots. Their pools are small, so one buy moves the price a lot, and their buyers tend to allow wide slippage because prices jump around. The deep pools of large coins barely show up.

We left out one kind of round trip. In a coin's first seconds, bots buy, a crowd of buyers piles in, and the bots sell into the rush. That is sniping, a separate behaviour, so round trips that start within about four seconds of a coin's first trade are not counted here. Round-trip bots made 3.2 million of these, worth $44.3 million to them, more than all the sandwiches together.

05 · The takeHow much do sandwich bots make?

The bots took 309,010 SOL from the trades they wrapped, worth $29.7 million at the price of SOL in the minute of each attack. That is the gross: what came back from the pools minus what went in.

Single attacks are small. The median sandwich made the bot $2.33, and one in ten made more than $12. Fees and tips take a cut. We checked 63 attacks on a public Solana node, 42 from one week in September and 21 from the other months, counting every fee and tip the bots paid. The bots kept 68% of the gross, and 18 of the 63 attacks lost money once costs were paid.

What the bots made came out of their victims' trades. On PumpSwap and on the launch curve the price follows a fixed formula, so for the sandwiches with one victim and nothing else in between we could work out how many coins each victim would have got without the bot. The victims' loss came to 1.3 times the bots' take on PumpSwap and 1.2 times on the launch curve. In those sandwiches, the take was a floor for what the victims lost.

Some of the take came out of other bots' trades. Weighted by what each victim paid, 39% of the take came from bot wallets such as snipers, and $18.1 million from everyone else.

06 · The botsWho runs the sandwich bots?

In all, 2,545 wallets made the sandwiches. The money is spread over about a thousand of them: the ten that took the most made 13% of it, and the top thousand made 95%.

The wallets do not last long. Half of them made sandwiches on 7 days or fewer, and no month saw more than 777 at work. Bot owners switch wallets often, which makes them hard to block one address at a time.

We looked for the first SOL that reached the 20 wallets that took the most, using Bitquery's transfer records up to 30 August. Of those, 11 got it from another wallet in that period, and three came straight from exchange hot wallets, the wallets exchanges pay customers from, at Bitstamp, Coinbase and Kraken, according to Bitquery's address labels. No funder paid for more than one of the 20. Of the other nine, six were already at work when our records begin, one started after the transfer records end, and two got no SOL from another wallet in the month before they started.

07 · SlippageDoes a high slippage setting invite an attack?

Jupiter, the swap router many Solana apps use, records the slippage limit of every swap it sends. In November last year, 21% of the wallets swapping through Jupiter allowed 10% or more on at least half of their swaps. By September that share was 30%, and in the first week of October 36%.

A wide setting leaves room for an attack. We matched every Jupiter swap against the sandwiches. Across all of them, swaps that allowed 20% to 50% landed inside a sandwich 96 times as often as swaps that allowed under 0.5%. Most of that gap is about which coins people traded. Nearly all the swaps with tight settings, 97% of those under 0.5%, went into pools other than Pump.fun's, where bots barely work.

So we compared swaps into the same kind of coin. Among Jupiter swaps into Pump.fun memecoins, those that allowed 20% to 50% were caught 14 times as often as those that allowed under 3%.

The tightest settings were not the safest. Swaps under 0.5% were caught more often than swaps at 1% to 3%, mostly in a burst in March and April, and swaps that allowed 50% or more were hit less often than those at 20% to 50%. Jupiter also carries only a small part of the victims' trades; most victims traded on Pump.fun directly. Past a few percent, the more room a trade leaves, the more often a bot takes it.

08 · Across blocksWhy do most attacks span several blocks?

In only 6% of the sandwiches did the bot buy and sell in the same block. Solana makes a block, which it calls a slot, about every 0.4 seconds, and each validator makes four in a row. In three of every four attacks the bot's sell came two to eight slots after its buy, so the two were often in blocks made by different validators.

That shape fits what the September study found. In 2024, sandwiches on Solana bunched up in blocks made by a few validators. The Solana Foundation and large staking pools then stopped sending stake to validators caught doing it, the last round in October 2025. The attacks did not stop. They moved to a form that needs no friendly validator, which suggests the bots learn about trades somewhere before the trades reach the chain. The study found the victims bunched up among users of a few trading apps, without saying how the orders leaked. We saw the same kind of leak on Robinhood Chain, where bots front-ran a trading app's memecoin orders.

09 · DefenceHow to avoid a sandwich attack

Keep slippage to a few percent. A wide setting such as 20% may save a failed trade during a rush, and our numbers show it is also the setting that bots catch most often.

Do not rely on a single protection switch. Many trading apps offer an MEV protection setting; MEV is the profit that bots and validators take by reordering trades. The most common switch only stops another trade from going ahead of yours inside a Jito bundle, a package of trades sent to a validator together. The September study found that most attacks no longer go that way, and that the most-hit app's victims had the switch on almost every time.

Bots make the most from large buys in small pools, where one order moves the price furthest. If you need a lot of a thinly traded memecoin, split the buy and keep each part's slippage tight.

10 · MethodHow we counted

We used Bitquery's records of every successful swap on the Solana exchanges it covers, from 30 October 2025 to 7 October 2026, 10.7 billion in all, on Pump.fun, PumpSwap, Raydium, Meteora, Orca and other pools. The same history is sold as files on the Bitquery Data Store and can be queried through Bitquery's DEX trades API.

A sandwich, by the September study's rules: a wallet buys a coin in a pool and later sells 99% to 101% of it back in the same pool, and the whole round trip fits inside three validators' turns in a row, about five seconds. In between, another wallet trades the same coin in the same direction, for at least 1% of the bot's buy and at a worse price than the bot got, and that wallet is not making such a round trip itself. A wallet counts as a sandwich bot if it made at least 100 sandwiches, at least 60% of them for a profit, and if sandwich trades were at least a quarter of all its swaps. We added one rule: round trips that start in a coin's first 10 slots are sniping and are left out, and the bot test is then run again without them. Counted with those round trips, as the study counts, our figure for November to June is 20,700 per day against the study's 20,663.

We checked the rule three ways. We took 78 sandwiches, 57 from one week in September and 21 from the other months, and checked each one on a public Solana node: the order of the three trades, the amounts, the pool, and that the bot's trades touched no other coin. All 78 held up. We wrote the rule a second time in plain code and ran it on the raw swaps of 25 sampled pools: it found the same 84 sandwiches, with the same victims. Then a separate program, written without our code, recounted the headline figures from the same records, and all 39 matched.

11 · EvidenceAddresses and transactions cited

FAQ

What is a sandwich attack in crypto?

A sandwich attack is when a bot buys a coin just before your trade and sells it just after. Your trade fills at the higher price the bot's buy created, and the bot sells at the higher price your buy created. The difference is the bot's profit and your loss.

How common are sandwich attacks on Solana?

Very common. We counted 6.9 million in the 11 months we studied, and the daily count in September 2026 was 15 times that of late 2025. Almost all hit memecoins on Pump.fun.

Who are the victims of sandwich attacks?

Mostly memecoin buyers on Pump.fun. Some are bots themselves, such as snipers, and their trades paid 39% of what the sandwich bots took. The rest came from other wallets.

How much do sandwich bots make on Solana?

About $29.7 million in our window, before costs. Most attacks made the bot a few dollars, but a few made far more.

What slippage should I set to avoid sandwich attacks?

A few percent at most. On Pump.fun memecoins, swaps that allowed 20% to 50% slippage got sandwiched far more often than swaps that allowed 1% to 3%. A wide setting saves a trade in a rush, but it is the room a bot needs.

Does MEV protection stop sandwich attacks on Solana?

Only partly. The most common protection switch stops other trades from going ahead of yours inside a Jito bundle, but a September 2026 study found most Solana attacks no longer go that way. Low slippage still matters.

Can a sandwich bot take coins out of my wallet?

No. A sandwich bot cannot touch your wallet. It only makes your trade fill at a worse price, within the slippage you allowed.