A Year of Fee Data: What We Learned

We sampled network fees at fixed times daily for a year. This is what the data shows.

The method

Base fee or equivalent recorded at 08:00, 14:00 and 20:00 UTC daily, across seven networks, converted to currency at the rate of the moment of sampling.

Three standard operations priced: a native transfer, a token transfer, and a swap through the most liquid venue on each chain.

Medians reported rather than means, because the distribution has a long right tail.

Finding one: the weekly cycle is stable and large

Weekend mornings are consistently the cheapest window on mainnet. Weekday afternoons in the European and North American overlap are consistently the most expensive.

The ratio between them averaged between three and five times across the year, and it did not weaken.

For any operation that can wait, this is the single largest available saving and it requires no skill.

Finding two: rollup fees decoupled and stayed decoupled

Since dedicated data capacity was introduced, rollup costs no longer track mainnet congestion in the way they previously did.

Rollup fees were both lower and substantially less variable than mainnet across the entire sampling period. The daily and weekly patterns visible on mainnet are barely detectable on rollups.

Finding three: the tail matters more than the average

Mean fees were pulled well above median by a small number of spike events, each lasting hours.

Anyone transacting without checking encounters those spikes at the same rate as everyone else. Anyone who checks avoids them entirely, which means the difference between a careful and a careless user is larger than the difference between any two networks in the normal range.

Finding four: swap costs and transfer costs diverge

Chains cheap for transfers are not always cheap for contract interaction, because execution and data are priced differently.

Comparisons using only transfers mislead anyone whose activity is mostly swapping, and most published comparisons use only transfers.

Finding five: venue withdrawal charges exceeded network costs on most platforms

Comparing published withdrawal fees against the actual network cost at the time showed a margin on most venues, of varying size.

Platforms that separate the network fee from their own charge, such as platforms that do not round the fee up, make this visible. Those quoting a combined figure do not.

The practical summary

Use a rollup for anything that does not require mainnet. Transact at the weekly low when timing is flexible. Check the current fee before anything discretionary. Compare venues on withdrawal charges rather than trading fees if you withdraw regularly.

Four habits, derived from a year of data, and none of them requires any judgement about markets.

Repeating the work

The method is above and the data sources are public. Anyone can reproduce it, and doing so on the specific networks and venues you actually use is more informative than our aggregate.