Trading terminal execution quality 2026: what a swap costs once the app's own fee is removed

As of the last measurement, Binance Wallet at 146 bps (p50, 24h) across 13 trading apps and Telegram bots. The figure is the median value lost per swap against the pool's state before the trade, with the app's own fee subtracted on each swap. It moves with every refresh; the board above is the current answer.

Not necessarily. An app with a high fee can route well and an app with no fee can route badly; this board isolates the routing. Compare a row's place here with its place on bench 268, trading terminal fill quality (/benchmarks/terminal-fill-quality): the gap between the two is the fee.

From the user's own transactions read on-chain. Each swap is priced against the pool's state before the trade, the shortfall is split into terminal fee, network, protocol and pool, and the terminal fee is removed. The median of what remains, over a rolling 24 hours, is the row's figure.

They answer different questions and a reader usually wants one or the other. Bench 268 is what a swap costs you, which is what you pay and includes the fee. This one is how well the app executes, which is what you would compare if you were choosing between apps whose fees you already know. The underlying swaps are identical and both boards publish the same rows under the same rules.

Because that gives the wrong number. The median of a difference is not the difference of the medians unless the fee is identical on every swap, which is false for any app running promotions, tiered fees or rebates. The size of the gap follows how much an app's fee varies across its own swaps; it is widest on an app like FOMO, which charges 2 % on Solana and about 0.5 % through Relay. So it moves with the cohort, and the board's two columns are the place to read it.

It stays in, because routing a trade through a launchpad curve that keeps 1 % rather than a migrated pool that does not is a decision the app makes. It is not a fee the app collects, but it is a cost the app chose for the user. The terminal, network, pool and protocol components are each published separately on bench 268 if you want to draw the line elsewhere.

No. It means the fee taken was larger than the total shortfall on that swap: a rebated or sponsored fee, or a reference price that drifted in the user's favour between the previous trade and this one. These are real swaps and stay in the sample; excluding them would bias every median upward.