Crypto Fee Statistics: What Using Crypto Has Cost, 2017–2026

On 20 April 2024, the average Bitcoin transaction cost $124.17. On 15 July 2026, the median transaction on Optimism cost $0.00001. Both figures are real. Crypto fees don't move in percentages — they move in orders of magnitude, and most numbers published about them are missing the qualifier that would make them true.

This page is the reference: what a transaction has actually cost, year by year, since 2017; the days the fee market broke; and what the 2024 blob upgrade did to the price of using a rollup. Everything below is computed from raw data, not copied from a press summary.

Method, stated up front

Two series do the work here. Bitcoin and Ethereum totals come from Coin Metrics' FeeTotNtv, defined as "the sum of all fees paid to miners, transaction validators, stakers and/or block producers that interval" — crucially, it counts fees that the protocol burns, so it measures what users paid, not what validators kept. Per-chain rollup costs come from growthepie's txcosts, which the project describes as "a median-style cost metric, not total cost". Days are UTC days. Bitcoin covers 4,579 days from 2014-01-01; Ethereum 4,004 days from 2015-07-30; both run through 2026-07-15.

One warning for anyone reproducing this: blockchain.info's fee chart is sampled when you request the full history — its 2017 series contains 91 points spaced every fourth day. Summing it gives $138M of Bitcoin fees for 2017 against a true $553.8M. It is reliable for single-day spot values and wrong for annual totals.

What a transaction has cost, by year

2026 covers 196 days to 15 July. At that pace the full year lands near $73.9M for Bitcoin and $179.0M for Ethereum — projections, not measurements.

Two things jump out. First, 2021 was the peak and nothing since has come close: Ethereum users handed over $9.92B in twelve months, more than double Bitcoin's $4.43B cumulative total since 2015. Ethereum's own cumulative figure, since its data begins on 2015-07-30, is $20.55B. Second, the collapse is not gentle. Ethereum's average fee per transaction in 2026 to date ($0.22) is 99% below 2021 ($21.48).

The days the fee market broke

Bitcoin's eight most expensive days, ranked by daily average fee per transaction:

The record day triangulates cleanly. CoinDesk reported that "Bitcoin transaction fees averaged a record $127.97 on April 20" citing BitInfoCharts, put single-day miner revenue at a record $107.8M citing YCharts, and attributed the "highest level ever of 75%" fee share per block to The Bitcoin Layer. Our figure is a full-day average on a different fee convention, which is why it reads $124.17 rather than $127.97, and 74.8% rather than 75%. Different conventions, same event.

On 2021 April, a correction is owed to the record. The cluster is routinely explained by a Xinjiang mining blackout crashing the hashrate. The article most often cited for that claim now carries an editor's note: "five-day stats show the Bitcoin (BTC) hashrate only lost around 25%" — the larger figure in its original text was withdrawn. Bitcoin's hashrate did hit an all-time high of 218 EH/s on 15 April 2021, and Cointelegraph reported individual pools down 18.9% to 33% while the overall network fell "roughly 2.2%" in 24 hours. Neither source mentions fees or the mempool at all. We publish the fee numbers and decline to assert the mechanism.

The same discipline applies to 2024-06-07, which is the second-highest fee day in Bitcoin's history and has no explanation we can source. Transaction count collapsed to 250,713 against 530–600k on neighbouring days while 322.7 BTC in fees were paid. We have found no account of why. The row stands without a cause.

December 2023 is worth a footnote for anyone checking our work against the archives. CoinDesk reported that the average fee "surged over $37 Sunday, a level last seen in April 2021." The comparison checks out precisely — scanning every prior day, the last one above $37 was 2021-04-26 at $37.85. The day does not: our peak that weekend is Saturday 2023-12-16 at $37.75, and Sunday computes to $36.24. Day-boundary conventions differ between providers, which is why we state ours.

Ethereum's five most expensive days:

The day before the Otherdeed mint, the average was $11.11. The mint multiplied it eighteenfold inside 24 hours. Published figures for that mint scatter wildly — Decrypt reported 55,843 ETH (~$157M) burned across the mint weekend, others quote $150M, $190M or 68,000 ETH. They disagree because each uses a different window and scope: seven hours, 24 hours, "the weekend." None is a clean UTC day. That is exactly why we publish the network-wide UTC-day figure of $231.3M and show our method.

Paid and earned are two different numbers

This is the most common error in fee reporting, and it has a birthday.

The two series track each other to within about 1.5% before London — and to the decimal on the days above — then split permanently on the fork date, 5 August 2021, block 12,965,000. The gap is the EIP-1559 burn. Etherscan's daily fee chart measures what validators are paid; Coin Metrics measures what users hand over. Neither is wrong. Quoting one while meaning the other is.

The size of the error, by year, taking Etherscan's tips against total fees paid: −38.2% (2021), −84.0% (2022), −83.5% (2023), −80.0% (2024), −53.1% (2025), −27.1% (2026 YTD). Anyone citing "Ethereum fee revenue" without saying which one they mean can be off by more than four fifths.

The reconciliation holds against an independent counter. Summing our implied burn from 2021-08-05 to 2026-07-15 gives 4,632,665 ETH; ultrasound.money's live burn total read 4,636,301.49 ETH at block 25,548,279 (2026-07-16, 22:40 UTC). That is 0.078% apart across five years and two independent data pipelines. Note that this is a ticking counter — cite it with a timestamp or it will drift out from under you.

Halvings don't do what people say they do

The standing thesis is that as the block subsidy shrinks, fees must rise to replace it. Four epochs of data say otherwise.

Cutting the subsidy in half between epochs 4 and 5 left the fee share essentially unchanged (5.86% → 5.71%). Cutting it again has so far seen the share fall by more than half. Epoch 3 is partial because our Bitcoin data starts in 2015; each halving day contains blocks from both epochs and is assigned whole to the new one, which moves nothing material.

A related claim needs pinning down. Scanning all 4,214 days from 2015-01-01 to 2026-07-15, exactly one had network-wide Bitcoin fees exceed the block subsidy: 2024-04-20, at 74.8%. The runners-up never got there — 2024-04-23 (45.9%), 2017-12-22 (43.6%), 2024-04-21 (43.4%), 2023-05-08 (42.3%). You will often read that fees first passed the subsidy in May 2023; that reporting was about individual blocks. CoinDesk's account says "mining pools such as Luxor Technologies and AntPool mined blocks on Monday where fees exceeded Bitcoin's block subsidy of 6.25 BTC." That is true of blocks and false of the day: 2023-05-08 closed at 42.3%. Blocks are not days.

What blobs actually did

EIP-4844, status Final, gave rollups a dedicated data lane — a target of three blobs (0.375 MB) per block and a ceiling of six (0.75 MB). It shipped with Dencun on 13 March 2024, block 19,426,587.

The obvious problem with measuring its effect is that fees fall for many reasons at once. So this table includes a control: Polygon PoS is a sidechain that posts no blobs and gained nothing from the upgrade. Both windows are 30 days.

The control moved 5.8%. The blob-posting rollups moved 77–99%. That difference is the upgrade, and it is the cleanest thing on this page. The two honesty guards matter: Ethereum L1's own decline is not a 4844 effect — L1 gained no blob lane, and its drop belongs to ordinary demand — and we have no sourced explanation for why Scroll barely moved.

Where costs sit now, as 30-day medians to 2026-07-15: Optimism $0.00001, Base $0.00119, Scroll $0.00186, Starknet $0.00232, Arbitrum $0.00404, Polygon PoS $0.00945, Linea $0.01193, Ethereum L1 $0.01355. Sub-cent is typical of major rollups, not universal — Linea, an L2, sits above a cent. The live per-chain readings are on the Crypto Fees Index, which currently scores 7.5as of 09:07 UTC.

Year by year, the same metric shows how far the floor fell:

— means the chain had not launched. Cells marked n= are partial years and are not comparable to full ones — Optimism's 2021 median rests on 51 days in a peak-gas window. Starknet 2021–2022 is excluded, not missing: growthepie reports a hard $0 for it before 2023, which we treat as unreported rather than free.

Ethereum L1's own median cost is down 99.84% from 2021 ($9.22683 → $0.01493). The major rollups are down three to five orders of magnitude. Optimism's 2021-to-2026 ratio works out to roughly 130,000× cheaper, though that number leans on those 51 days and should be handled accordingly.

Fees by era

The through-line is that fees have never been a property of a chain. They are a property of a moment — an auction for block space that a single NFT mint or a protocol launch can move by two orders of magnitude in a day, and that years of quiet demand can grind back down to a hundredth of a cent. Any fee statistic without a date attached is decoration.