Over an 11-month span, sandwich bots pulled roughly $29.7 million in gross value from Solana trades, according to data from blockchain analytics provider Bitquery. The research revealed that 96% of the identified attacks targeted Pump.fun’s launch curve and its PumpSwap exchange.
Analyzing successful swaps between October 30, 2025, and October 7, 2026, the study documented 6,850,144 attacks executed by 2,545 bot wallets, impacting 2,154,994 unique wallets. These addresses do not equate to individual people, as trades originating from other bots made up 39% of the estimated gross earnings.
A sandwich attack occurs when a bot places a buy order and a sell order directly around another user’s swap. The initial purchase drives up the price in the liquidity pool, forcing the intervening buyer to acquire fewer tokens, after which the bot liquidates its holdings at the elevated price. The financial impact is absorbed by the trader during execution, even when the swap finishes successfully.
Wide slippage marked greater exposure
Bitquery discovered that among swaps routed through Jupiter onto Pump.fun’s launch curve or PumpSwap, transactions permitting a 20% to 50% slippage tolerance were sandwiched 14 times more frequently than those capped below 3%.
Slippage tolerance defines the maximum permissible variance between an expected quote and actual execution. Documentation from Jupiter explains that this parameter establishes a minimum acceptable output once the tolerance is applied. If price movements exceed this threshold, the swap fails, whereas a wider limit allows for an unfavorable fill. This creates a compromise between successfully executing a volatile transaction and accepting a poorer price along the way.
The observed frequency of attacks did not scale uniformly with every broader setting. Tolerances set below 0.5% experienced more attacks than the 1% to 3% bracket, while the band of 50% or higher was targeted less frequently than the 20% to 50% range.
This slippage comparison is observational and focuses strictly on Jupiter routes that utilized Pump.fun’s launch curve or PumpSwap. Furthermore, Jupiter accounted for only a minor fraction of all victimized trades included in the study.
The $29.7 million figure calculates SOL values at the precise minute of each attack and does not reflect net profit. In a separate fee sample of 63 attacks analyzed by Bitquery, bots kept 68% of total gross proceeds, while 18 attacks resulted in a net loss after accounting for expenses. That sample does not determine net earnings for the entire period. Additionally, the report notes discrepancies in how Pump.fun sale fees were recorded, which alters the gross estimate by roughly 3%.
To identify the bots, Bitquery relied on recurring trading behaviors, omitted round trips that commenced within a token’s first 10 slots, and noted occasional gaps in exchange and block data coverage. On-chain records cannot definitively prove that a bot viewed an order beforehand, meaning certain qualifying wallets might simply belong to fast traders.
While slippage limits govern the price a trader is willing to accept, a separate research preprint submitted on September 23, 2026, investigated protected order flow—transactions intended to avoid front-running—across six blockchains. The authors documented sandwich patterns spanning multiple trades and blocks, leading them to conclude that these protective measures may provide weaker safeguards than users anticipate.
Frequently Asked Questions
How much money did Solana sandwich bots extract?
Sandwich bots extracted about $29.7 million in gross value from Solana trades over an 11-month period, according to data from Bitquery.
Which platform was targeted the most by these bots?
The vast majority of detected attacks—96%—targeted the memecoin platform Pump.fun’s launch curve and its PumpSwap exchange.
Did wider slippage settings increase the risk of an attack?
Yes, trades allowing 20% to 50% slippage were sandwiched 14 times as often as those allowing less than 3%, though attack rates did not rise uniformly across every wider setting.
Does the $29.7 million figure represent net profit for the bots?
No, the estimate values SOL at the exact minute of each attack and does not measure net profit. A smaller sample indicated that bots retained 68% of gross proceeds after fees, with some individual attacks losing money.
