Solver auctions: inside the $23.3 billion market that settles your swap

01 · What happens after you press swap

Your trade is an auction, and solvers are the bidders

CoW Swap is a DEX. You sign an intent: what you will give, and the least you will take back. Solvers bid to fill it. The winner packs the order into a batch and pays the gas. A pool never sees it.

On Ethereum, roughly 30 firms each compute a route. The best bid wins. The same loop runs every few seconds on five chains.

The order never sits in the public mempool, so nothing can sandwich it. Solvers take the spread between what you signed for and what the market gives, plus a COW reward for every win.

Bitquery indexes every call and event on Ethereum, BNB Chain, Arbitrum, Base and Polygon. We pulled the 2026 record for CoW Protocol and its three rivals, UniswapX, Bebop and Relay, to see how big this market is and who takes the pay.

02 · Five chains, one market

BNB Chain carries the trades now

CoW settles the same way everywhere: one contract, same address on all five chains, one auction per batch. Where those auctions fire has changed this year.

BSC went from a rounding error in January to the busiest venue by July, with the wallets spread out across the chain. Base arrives already large, three and a half months of data, and already settles at Arbitrum's level. Polygon barely moved in September, and its ten biggest solvers take nearly all of the batches.

03 · The dollars stay on Ethereum

Count the trades and BSC wins. Count the money and it is not close

Line the Ethereum orders up from smallest to largest and the one in the middle, the median, is $734. Half of mainnet flow is retail-sized. A thin layer of whales pulls the mean about 37 times higher, which is why the average looks like a different market. On BSC the typical ticket is a few hundred dollars.

Gas is why the cheap chains took the flow. Settling on Ethereum costs a solver over a dollar per batch. On Arbitrum the same job costs two cents.

Cheap gas also changes how solvers pack a batch. On Ethereum and BSC a settlement holds about one order, so the auction is mostly one trade at a time. On Arbitrum a batch holds one and a quarter. Many orders crossing each other in one tx is still rare.

CoW's Ethereum volume is a thin slice of the roughly $280 billion of mainnet DEX volume in the same window, something like seven or eight cents of each mainnet dollar.

04 · The solvers

High churn, and one team with no name

Thirty-two solver addresses won auctions on Ethereum in the first half of this month. Public dashboards name most of the teams. Those names are the addresses written into each settlement event.

Year totals hide the churn. BRRRolver is tenth on the year board and has been taking a larger share of new auctions on Ethereum. Tsolver and Baseline went the other way. The top five on Ethereum win under half of all auctions, which makes mainnet the most open venue. On BSC the top five take three quarters. On Arbitrum they take about seven in ten. On Base, two thirds.

Base's busiest solver, with a third of every settlement there since June, has no public name. It does not appear in CoW's own solver lists. Whoever runs it showed up on the newest chain and took the top slot.

05 · What the job pays

36.8 million COW, and half of it to one team

CoW pays solvers for each win out of a known rewards safe, in transfers anyone can read.

The COW price comes from a median of two on-chain trades, so treat the dollar figures as rough. Rewards are the visible part of solver pay. Surplus kept inside a winning batch never shows up as a transfer. The ledger shows about $3.6 million paid out against roughly $1.0 million of gas burned. The gas refunds nearly cover the mainnet gas bill.

Failed settle calls are a rounding error on mainnet, about one in 500. On Base, the worst venue, about one in 65. In a three-day September sample, the median WETH-to-USDC order priced within five basis points of the same-block Uniswap median. The auction matches spot. What the trader gets is MEV cover and a fill that never sat in the mempool.

06 · Who uses it

Two thirds trade once and vanish. One percent are machines

BSC is less concentrated on the same test. In July, 67,000 wallets traded and the ten busiest took a fifth of orders.

About 7% of June's CoW orders on BSC were sales of TUFT, a token tied to a pig-butchering scam. Victims were told to swap it back to USDT, and CoW handled the exits: about 24,000 orders this year, 3% of the chain's flow, traced with money-flow tools.

07 · The rival that left the chain

1inch Fusion's settlement contract has not traded all year

UniswapX, Bebop and Relay all run auctions of their own. Relay's fills are settled by a single in-house solver.

1inch Fusion recorded zero transactions on Ethereum and on BSC this year. We checked twice, on two replicas. Fusion still has a website and docs. On-chain, the auction that once competed with CoW for Ethereum order flow has gone quiet. Where that flow went, the ledger does not say.

08 · How we measured it

The universe, and one count we had to fix

The counts come from the settlement contract CoW deploys at the same address on all five chains. A settlement is its Settlement event, an order is its Trade event, a trader is the order's owner, and a solver is the address the event names as the winner. The window is January through mid-September. Every figure was run twice on separate replicas. Duplicate-event scans and cross-checks between the archive and realtime indexes sit behind the numbers.

The decoded event record on Base is missing for a year before June, so Base covers three and a half months. The raw chain record is complete, and a four-day fragment from May 2025 proves CoW was live on Base a year before our window opens.

Volume is a floor. We count the sell side in major tokens at prices from the dates traded, and Polygon's figure is a 30-day snapshot. The intent-share estimate uses an outside benchmark for mainnet DEX volume, since raw trade amounts in any index carry decoder outliers. Execution quality was checked against same-block Uniswap v3 medians for the busiest pair, a proxy rather than a tick-level audit.

One number failed audit and was fixed before print. Our first Base settlement count came back twenty-seven percent low because of a bad filter; the corrected read, verified three ways on two replicas, is what you see above. As an outside anchor, CoW DAO's own August recap reports $2.88 billion for the month across every chain it supports; our majors-only August read sits just under it, as a lower bound should. All of it was pulled with Bitquery's trade API and price feeds.