Sandwich bots on Solana extracted roughly $29.7 million before costs from other people's trades over an 11-month stretch, according to blockchain data provider Bitquery. The firm counted 6,850,144 attacks carried out by 2,545 bot wallets, hitting 2,154,994 distinct wallets between October 30, 2025, and October 7, 2026.
The overwhelming majority of those attacks — 96% — landed on Pump.fun's launch curve and its PumpSwap exchange. The $29.7 million figure values SOL at the minute of each attack. It is not a net profit number.
Where the bots hunted
Bitquery's breakdown points at a specific weakness in how slippage tolerance gets set on Pump.fun-adjacent trades. Among swaps routed through Jupiter that touched the launch curve or PumpSwap, trades allowing 20% to 50% slippage were sandwiched 14 times as often as those allowing under 3%.
The pattern isn't a clean ladder, though. Trades with under 0.5% tolerance got hit more often than those in the 1% to 3% band. The 50%-or-more band was hit less often than the 20% to 50% band. That unevenness matters: it means "looser setting equals more attacks" isn't the whole story.
Jupiter carried only a small share of all victim trades in the investigation, so its numbers describe a slice, not the market.
What the number does and doesn't cover
Bitquery used repeated trading patterns to flag wallets as bots. It excluded round trips that began in a token's first 10 slots, and it reports gaps in exchange and block coverage — both caveats that could cut either way on the total.
There's also a fee question. In a separate 63-attack sample, bots kept 68% of aggregate gross proceeds. Eighteen of those attacks lost money after costs. Bitquery is explicit that this sample does not establish net earnings for the full period, so the headline $29.7 million shouldn't be read as take-home pay.
An inconsistent recording of Pump.fun sale fees throws off the gross estimate by about 3%, per the report. And other bots' trades accounted for 39% of the estimated gross take — meaning a chunk of the value moved wasn't extracted from human users at all. The 2.15 million affected addresses are addresses, not people.
The limits of on-chain forensics
This is the part that keeps sandwiching cases hard to prosecute or even to describe cleanly. Chain records can't prove a bot saw an order before it placed its own. Some wallets that qualify under Bitquery's method may just be fast traders. The firm flags both possibilities rather than resolving them.
A separate research preprint submitted September 23, 2026 looked at protected order flow across six blockchains. It found sandwich patterns spanning across trades and blocks, and concluded that these protections can offer weaker guarantees than users expect. That's a broader warning than Solana alone.
What comes next
Bitquery hasn't published a net-profit figure for the 11-month window, and the fee-recording gap on Pump.fun leaves the gross estimate with a stated error bar. Until someone closes that gap — or the preprint's cross-chain findings get tested against live order-flow systems — the $29.7 million is best treated as a gross extraction estimate with known holes in it.
