Research

Measured crypto payment transaction fees, by year, 2009 to 2026

13,635,109 blocks measured, 2009 to 2026 · updated July 29, 2026

Measured on our own node across every Bitcoin block from height 1 to 960,129: the median fee in 2026 was 160 satoshis at 1 sat/vB, against 41,174 at 134 sat/vB in 2017, the costliest year for block space. Ethereum's median base fee in 2026 was 0.1136 gwei, a floor of 0.00000239 ETH per transfer.

Bitcoin (BTC)

every block from height 1 to 960129 · 960,129 blocks

Year	Blocks	Transactions	Median fee	Mean fee	Median fee rate	p10 – p90 fee rate
2009	32,489	219	0 sat	1,310,502 sat	0 sat/vB	0 – 0
2010	67,920	117,385	0 sat	37,475 sat	0 sat/vB	0 – 0
2011	59,627	1,842,138	0 sat	167,551 sat	0 sat/vB	0 – 193
2012	54,526	8,398,524	50,000 sat	80,936 sat	62 sat/vB	0 – 134
2013	63,433	19,579,808	50,000 sat	78,012 sat	76 sat/vB	11 – 173
2014	58,865	25,204,855	10,000 sat	18,396 sat	20 sat/vB	11 – 38
2015	54,321	45,619,702	10,000 sat	17,975 sat	19 sat/vB	8 – 32
2016	54,851	82,571,772	12,434 sat	27,316 sat	40 sat/vB	12 – 65
2017	55,928	104,007,301	41,174 sat	96,504 sat	134 sat/vB	25 – 350
2018	54,498	81,341,138	2,289 sat	31,601 sat	8 sat/vB	2 – 100
2019	54,232	119,729,415	3,969 sat	16,527 sat	16 sat/vB	2 – 80
2020	53,222	112,500,276	6,256 sat	23,386 sat	21 sat/vB	3 – 109
2021	52,686	97,743,767	2,712 sat	21,957 sat	10 sat/vB	2 – 103
2022	53,188	93,053,190	1,667 sat	5,776 sat	6 sat/vB	1 – 17
2023	53,999	153,361,994	3,255 sat	15,279 sat	15 sat/vB	3 – 88
2024	53,473	192,195,135	1,733 sat	7,904 sat	10 sat/vB	2 – 45
2025	53,082	153,679,189	356 sat	1,119 sat	2 sat/vB	0 – 4
2026	29,789	115,151,576	160 sat	507 sat	1 sat/vB	0 – 3

Fees are in satoshis (1 BTC = 100,000,000 sat) and fee rates in satoshis per virtual byte. We do not convert to fiat: that would require a third-party historical price series, which is exactly the kind of secondary source this dataset exists to avoid.

Ethereum (ETH)

every block from the London fork (12965000) to 25639979 · 12,674,980 blocks

Year	Blocks	Median base fee	p10 – p90 base fee	Transfer floor	Median gas used
2021	951,166	81.04 gwei	40.16 – 150.95	0.001702 ETH	0.4629
2022	2,392,024	22.44 gwei	8.18 – 94.29	0.000471 ETH	0.4628
2023	2,600,705	22.07 gwei	11.05 – 50.06	0.000463 ETH	0.4661
2024	2,616,996	10.3 gwei	1.98 – 37.05	0.000216 ETH	0.4757
2025	2,610,162	0.5527 gwei	0.0883 – 4.59	0.00001161 ETH	0.4734
2026	1,503,927	0.1136 gwei	0.0378 – 0.466	0.00000239 ETH	0.4775

Figures are EIP-1559 base fee per gas, in gwei. Base fee is the protocol-mandated floor that is burned; it excludes the priority tip a sender adds to get included, so treat every number here as a lower bound on what a transaction actually cost.

Methodology

GriffNode operates its own full archival Bitcoin node (txindex enabled, no pruning) and its own Ethereum execution node (Erigon). For Bitcoin we called getblockstats on every block from height 1 to the chain tip; Bitcoin Core derives those figures from the block and its undo data, so the fee of every transaction in every block is accounted for. For Ethereum we read the baseFeePerGas field from the header of every block from the London fork onward. Blocks are bucketed into calendar years by their own UTC timestamp. Within a year, the median fee is the median across blocks of each block's own median transaction fee, so every block counts once regardless of how many transactions it carried; the mean fee is instead the year's total fees divided by the year's total non-coinbase transactions, so it is transaction-weighted. Both are reported because they answer different questions and diverge sharply during congestion. Percentiles use linear interpolation. No smoothing, outlier removal or clamping is applied. The collection script is published alongside this dataset and re-runnable against any full node.

Caveats

The Bitcoin median is a median of per-block medians, not a median over every transaction ever confirmed. Weighting every block equally rather than every transaction equally is a deliberate choice: it stops a handful of enormous blocks during congestion from dominating a whole year. The transaction-weighted mean is published in the same row precisely so the two can be compared, and a large gap between them is itself the signal that a year was congested.
The Ethereum series is base fee only. EIP-1559 splits what a sender pays into a burned base fee, which is in the header and which we measure exactly, and a priority tip paid to the proposer, which is not in the header. Every Ethereum figure here is therefore a lower bound on the true cost of a transaction, and the gap widens during periods when block space was contested.
Ethereum coverage starts at the London fork (block 12,965,000, August 2021) and cannot be extended backwards from this node. Our Erigon instance runs with --prune.mode=full, so historical transactions and receipts are not on disk; pre-London fees would require an archive node we do not operate. We would rather state that limit than fill the gap with a third-party series and present it as our own measurement.
Fees are reported in each chain's native unit -- satoshis and sat/vB for Bitcoin, gwei for Ethereum -- and never converted to fiat. A fiat conversion would depend entirely on a historical exchange-rate series published by someone else, which would make the resulting number a secondary source wearing a primary source's clothes.
A sat/vB fee rate is a price per unit of block space, not a price per payment. What a given payment costs also depends on how many inputs it spends, so a merchant consolidating many small deposits pays more at the same fee rate than one spending a single input. Read the fee-rate columns as the market price of block space and the fee columns as what a typical transaction of that era actually paid.
Calendar-year buckets are a reporting convenience, not a natural unit for fee markets, which move on halvings, forks and demand shocks. A year containing a single severe congestion episode and eleven quiet months will show a median that describes neither.
The earliest Bitcoin years are a curiosity, not a fee market, and the 2009 row shows why: across the whole year only eight blocks carried any fee at all, 219 non-coinbase transactions paid 2.87 BTC between them, and a single block at height 2,817 accounted for 2.01 BTC of that from four transactions. So the median is genuinely 0 (almost nothing paid) while the mean is enormous (a handful of hand-set fees dominate). Both figures are correct and neither describes a market. The blocks_with_no_paying_tx column shows how many blocks were excluded from each year's median for carrying no paying transaction.

Use the data

Published under CC BY 4.0: quote it, chart it, cite it. The full table is available as raw CSV ↓

GriffNode. "Measured crypto payment transaction fees, by year, 2009 to 2026". 2009-2026. https://learn.griffnode.com/research/crypto-transaction-fees (CC BY 4.0). Retrieved July 29, 2026.

Built on the same nodes that settle your payments

GriffNode is a non custodial crypto payment gateway. Funds go straight to your wallet, confirmed by the nodes that produced this data.

Get started free →