Ethereum gas fees have swung by a factor of almost a thousand for the same job. On 7 December 2017, the week CryptoKitties clogged the network, sending ether from one wallet to another cost about 53 US cents. On 1 May 2022 the same transfer cost about 28 dollars. On 9 November 2026 it cost about 3 cents. Those figures come from Etherscan’s daily average gas price multiplied by the ETH price of each day, so they are averages, not receipts. The work did not change once: a plain ether transfer uses 21,000 units of gas, then and now. Only the price of each unit moved.
That split between the amount of work and the price of work is the whole story of Ethereum gas fees, which our glossary entry on gas fees defines in two lines. Gas is a price list: every operation of the Ethereum Virtual Machine has a fixed cost in gas units. EIP-1559, live since the London upgrade of August 2021, is the rule that sets what one unit costs in each block, and it sends most of that money nowhere at all: it is burned. Both halves are taken apart below with real mainnet blocks from 10 November 2026, a fee worked out by hand, the burn arithmetic and the blob market that now carries most layer-2 data.
Two earlier lessons give the basics. Our guide to how crypto transactions work already explains that gas measures work, that a plain transfer uses 21,000 units and that the wallet sets a maximum fee and a tip. The previous lesson of this path, how the Ethereum Virtual Machine works, shows the machine that consumes the gas, opcode by opcode. This lesson starts where they stop, at the formula behind the price.
If you came with one question, here is the short answer. Ethereum fees were high, sometimes very high, from 2017 to 2022, because block space was scarce and demand came in bursts. As of 10 November 2026 they are not: the base fee sat near 0.06 gwei that morning and a plain transfer cost a fraction of a cent to a few cents. What still costs real money on Ethereum is heavy work, such as deploying a contract or writing many storage slots, on a busy day.
Why Ethereum gas fees exist and how they are computed
A meter against spam and code that never stops
Ethereum lets anyone run code on thousands of computers at once. Without a price on that work, one person could flood the network with useless transactions, or deploy a loop that never ends and freeze every node that tries to run it. The EVM lesson covers the second problem in detail: since no program can tell in advance whether arbitrary code halts, Ethereum does not try. It meters every step instead, and the transaction stops when its gas runs out. The original whitepaper already had this design, with a STARTGAS and a GASPRICE field, as our history of Ethereum and its whitepaper recounts.
Two quantities are therefore at play. Gas units measure work and are fixed by the protocol for each operation. The gas price says how much ether you pay per unit, and it is quoted in gwei. One gwei is a billionth of an ether (10-9 ETH); the smallest unit, the wei, is 10-18 ETH. A base fee of 0.06 gwei means 60 million wei per unit of gas.
Fee = gas used x price per unit of gas
The fee of any transaction is the gas it actually used multiplied by the price it paid per unit. The ethereum.org documentation is blunt about one consequence: “The fee is paid regardless of whether a transaction succeeds or fails.” A transaction that reverts halfway has still made every node do the work up to that point, so the work is billed.
Before a transaction runs, the sender must hold enough ether to cover the worst case: the gas limit they set multiplied by the maximum price they accept. That amount is reserved up front. Whatever is not used comes back at the end, which is why a generous maximum costs nothing extra on a quiet day.
What common operations cost in gas units
Each EVM operation has a fixed gas cost. The cheapest arithmetic costs 3 units. Touching storage, the permanent memory of a contract, is by far the most expensive thing a transaction can do, because every node must keep that data forever. The table below lists the costs that explain most real bills, from the protocol’s execution specifications and EIP-2929; the opcode-by-opcode tour is in the EVM lesson.
| Operation | Gas units | Why it matters for your fee |
|---|---|---|
| Any transaction (intrinsic cost) | 21,000 | The floor: a plain ETH transfer uses exactly this |
| ADD, SUB, PUSH, DUP, SWAP | 3 | Arithmetic is almost free |
| MUL, DIV | 5 | Same |
| First read of a storage slot (cold SLOAD) | 2,100 | Token balances live in storage |
| First touch of another account (cold access) | 2,600 | Calling a token or a pool |
| Later reads of the same slot or account (warm) | 100 | Repeat access is cheap |
| Write a storage slot from zero to non-zero | 20,000 (plus 2,100 if cold) | Why a first token receipt costs more |
| Create a contract (CREATE base) | 32,000 | Plus 200 per byte of deployed code |
| Emit a log (LOG0), per topic, per data byte | 375 / 375 / 8 | Events such as Transfer |
| Calldata per non-zero / zero byte | 16 / 4 (floor 40 / 10 since Pectra) | Data sent with the transaction |
Constants do not tell you what a real transaction costs, so Coindaro sampled ten mainnet blocks on 10 November 2026 (every fourth block from 26,383,670 to 26,383,706, about 09:45 UTC) and read the gas used by each successful transaction from its receipt. The medians are steady and match what wallets show.
| Kind of transaction (Coindaro’s sample of 10 blocks, 10 Nov 2026) | Transactions | Median gas used | Middle half of the sample |
|---|---|---|---|
| Plain ETH transfer | 796 | 21,000 | 21,000 |
| ERC-20 token transfer (mostly USDT and USDC) | 1,114 | 45,160 | 41,297 to 58,373 |
| ERC-20 approval | 58 | 48,549 | 46,433 to 49,083 |
| transferFrom (ERC-20 or NFT) | 40 | 48,613 | 43,706 to 53,919 |
| Uniswap Universal Router swap | 53 | 120,523 | 100,476 to 148,199 |
A token transfer costs about twice a plain transfer because it calls a contract, reads two balances from storage and writes both back, then emits a Transfer log. A Uniswap blockchain, where users swap tokens directly from their own wallets without handing custody of their funds to a company.">swap costs five to six times a plain transfer: it touches the router, one or more pools and two tokens, reads and writes several storage slots and emits several logs. At that morning’s base fee of 0.0639 gwei, the median swap used 120,523 x 0.0639 = about 7,700 gwei, or roughly 1.6 US cents at an ETH price of 2,097.46 dollars (Etherscan, 09:42 UTC), before any tip. Etherscan’s own gas tracker estimated a swap at about 4.6 cents at the same moment; its model is not published on the page, so treat both as orders of magnitude.
One detail since the Pectra upgrade of 7 May 2025: EIP-7623 added a floor price for calldata, 40 gas per non-zero byte instead of 16 when a transaction carries a lot of data and does little computation. It targets transactions that used Ethereum as a cheap data dump; ordinary transfers and swaps are not affected.
Gas limit of a transaction, gas limit of a block
The word “gas limit” means two different things. In your wallet it is the most gas your transaction may use; if execution needs more, it stops, reverts and still pays for what it consumed. For a plain transfer wallets set it to 21,000 exactly. For contract calls they estimate it and add a margin.
The block gas limit is the most gas all transactions in one block may use together. London set it at 30 million in August 2021, with a target of 15 million. It stayed there for years, then validators started to raise it. Each block can move the limit only by a small step from its parent, and validators choose the direction through a setting in their client software, so no hard fork is needed. Block headers read by Coindaro show the climb: about 36 million on 4 February 2025, 45 million on 24 July 2025 and just under 60 million on 30 November 2025. The Fusaka upgrade of 3 December 2025 then wrote 60 million into the clients as the default (EIP-7935); it did not raise the limit itself, which validators had already done. On 10 November 2026 most blocks carry a limit of exactly 60,000,000 gas, with a target of half that, 30 million; a few proposers still sit a hair below, such as block 26,383,717 at 59,941,408. That small spread is the signalling at work.
Fusaka also added a cap on single transactions: since EIP-7825, no transaction may use more than 16,777,216 gas (224). A block can hold 60 million, but no one transaction can take it all. Ordinary users never come near that cap; it matters for large contract deployments and for keeping block validation predictable.
Before EIP-1559: a blind auction for block space
Until August 2021, each transaction named a single gas price and miners picked the highest bids. It was a first-price auction: you paid what you offered, not what was needed. Nobody knew the right price in advance, so wallets guessed, users overpaid on calm days and got stuck on busy ones. Bots that needed to be first, for arbitrage or liquidations, bid against each other in public, a pattern called the priority gas auction that our guide to MEV describes.
Bitcoin still works this way, with fees quoted in satoshis per virtual byte; our lesson on the mempool and transaction fees explains that market and gives a first look at Ethereum’s. The contrast is useful: Bitcoin prices bytes and lets miners keep every fee, while Ethereum since London prices computation, posts a base price per block and burns it.
EIP-1559: a posted price, a tip and a burn
EIP-1559 was proposed on 13 April 2019 and went live with London at block 12,965,000, mined on 5 August 2021 at 12:33:42 UTC. It replaced the single bid with three numbers: a base fee set by the protocol, a priority fee or tip chosen by the sender, and a maximum fee the sender will not exceed. The specification states the most radical part in one line: “The base fee is always burned (i.e. it is destroyed by the protocol).”
The base fee: one price per block, set by a formula
Every block has a single base fee that every transaction in it must pay per unit of gas. It is computed from the parent block alone. If the parent used exactly the target (half the gas limit), the base fee stays the same. If it used more, the base fee rises; if less, it falls. The change is proportional to the gap, divided by 8, which is why ethereum.org says the base fee “will increase or decrease by a maximum of 12.5% per block”. A completely full block, at twice the target, raises it by exactly 12.5 percent; an empty block cuts it by 12.5 percent.
In integer arithmetic, as clients compute it: new base fee = parent base fee + parent base fee x (gas used – target) / target / 8, with the division rounded down, and at least 1 wei when the block is above target. The first London block started at 1 gwei, the constant INITIAL_BASE_FEE. That first block was almost full, 30,025,257 gas against a 15 million target, so the next one opened at 1,124,967,822 wei, a hair under 1.125 gwei and almost the formula’s maximum step.
Here are five consecutive mainnet blocks from 10 November 2026, with a gas limit of 60 million and a target of 30 million. Recomputing each next base fee with the formula gives the exact number the chain recorded.
| Block (10 Nov 2026) | Base fee (wei) | Gas used | Versus 30M target | Next block’s base fee (wei) | Change |
|---|---|---|---|---|---|
| 26,383,712 | 64,849,735 | 16,844,723 | 56% | 61,295,085 | -5.5% |
| 26,383,713 | 61,295,085 | 55,296,829 | 184% | 67,755,798 | +10.5% |
| 26,383,714 | 67,755,798 | 31,531,443 | 105% | 68,188,148 | +0.6% |
| 26,383,715 | 68,188,148 | 35,521,789 | 118% | 69,756,983 | +2.3% |
| 26,383,716 | 69,756,983 | 25,713,230 | 86% | 68,511,016 | -1.8% |
Take block 26,383,713 by hand. It used 55,296,829 gas, 25,296,829 above target. The step is 61,295,085 x 25,296,829 / 30,000,000 / 8 = 6,460,713 wei, rounded down. Add it to the parent’s base fee and you get 67,755,798 wei, the base fee of the next block. The same check works on any block since London.
The rule is gentle block by block and fierce over minutes. A computed example, not live data: start at 100 gwei and fill ten blocks in a row. The base fee goes to 112.5, 126.6, 142.4, 160.2, 180.2, 202.7, 228.1, 256.6, 288.7 and 324.7 gwei. It doubles after six full blocks, about 72 seconds at 12 seconds per block, and more than triples in two minutes. The reverse is as quick: six empty blocks cut it by more than half. One crowded block barely moves the price; a sustained rush multiplies it.
This is what “elastic blocks” means. A block aims for half its limit but may use up to all of it, as block 26,383,713 did with 184 percent of target. Demand spikes are absorbed by bigger blocks for a moment, then priced away by the rising base fee, instead of leaving transactions to bid blindly against each other.
Max fee, tip and refund on a real transaction
A type-2 transaction, the EIP-1559 format, carries two price fields. The max fee per gas is the most you accept to pay per unit in total. The max priority fee per gas is the tip you offer the validator who includes you. The price actually paid is the block’s base fee plus the tip, capped by the max fee. If your max fee is below the block’s base fee, the transaction cannot enter that block and waits.
A live example, chosen for this lesson: transaction 0xde86…8c8a, a plain ether transfer in block 26,383,717, mined on 10 November 2026 at 09:48:59 UTC. On its Etherscan page the fields read as follows, and the arithmetic below them is ours.
| Etherscan field | Value shown | What it means |
|---|---|---|
| Txn Type | 2 (EIP-1559) | Base fee plus tip format |
| Gas Limit & Usage by Txn | 21,000 | 21,000 (100%) | A plain transfer uses exactly its limit |
| Gas Fees: Base | 0.068511016 gwei | Set by the protocol for block 26,383,717 |
| Gas Fees: Max | 0.235 gwei | The sender’s ceiling per unit |
| Gas Fees: Max Priority | 0.01 gwei | The tip offered |
| Gas Price | 0.078511016 gwei | Base 0.068511016 + tip 0.01 |
| Transaction Fee | 0.000001648731336 ETH (about $0.0035) | 21,000 x 0.078511016 gwei = 1,648.73 gwei |
| Burnt | 0.000001438731336 ETH | 21,000 x the base fee: 87.3% of the fee |
| Txn Savings | 0.000003286268664 ETH | Reserved at the max fee, then refunded |
The sender reserved 21,000 x 0.235 = 4,935 gwei up front, paid 1,648.73 gwei and got 3,286.27 gwei back: exactly the “Txn Savings” line. Of the fee paid, 1,438.73 gwei were burned and 210 gwei went to the block proposer as the tip. The whole transfer cost about a third of a US cent. For a full tour of every field on a transaction page, see our guide to reading a block explorer.
Tips in 2026 are tiny. Two other plain transfers in the same block tipped 0.0224 and 0.002134 gwei, and the public node we queried suggested a default tip of 22 wei, effectively zero. Etherscan’s gas tracker at 09:42 UTC showed a priority fee of 0 for its “Standard” level and 0.006 gwei for “Rapid”. A tip matters only when blocks are full and validators must choose; the rest of the time the base fee does all the work.
Do not confuse this refund with the “gas refunds” in the protocol. Clearing a storage slot used to earn back a large share of gas, which let a trick called GasToken store gas when cheap and spend it later. EIP-3529, also part of London, cut that refund and capped all such refunds at a fifth of the gas used. The max-fee refund above is a different thing: simply the unused part of your reservation.
The burn: where the base fee goes
The base fee is not paid to anyone. Validators get only the tips, plus the separate staking rewards of proof of stake. The ether paid as base fee leaves circulation. Before London, miners kept the whole fee, which gave them a reason to fill blocks with their own transactions to push fees up; burning the base fee removes that incentive and makes the price harder to manipulate.
The sums are large. According to the ultrasound.money data feed, read on 10 November 2026 at 09:42 UTC, 4,643,566 ETH had been burned since London, about 3.8 percent of today’s supply, worth 13.12 billion dollars at the price of each day it was burned. The record single event came during the Otherside land sale by Yuga Labs in spring 2022: Decrypt reported on 2 May 2022 that the mint burned 55,843 ETH, worth about 157 million dollars, in a weekend. Etherscan’s chart shows 1 May 2022 as the highest daily average gas price of that year, 474.57 gwei.
Ultrasound money: the idea and the arithmetic
“Ultrasound money” is the community’s nickname for the hope that the burn would exceed new issuance, so that the ETH supply would shrink. The Defiant, in a piece of 9 July 2021, credits Ethereum Foundation researcher Justin Drake with launching the meme in a single tweet. The idea needed two things: fees high enough to burn a lot of ether, and issuance low enough to be outburned. The second arrived with The Merge on 15 September 2022, when proof of stake replaced mining and cut new issuance sharply. The first did not last.
| Supply figure (ultrasound.money data, 10 Nov 2026) | Value |
|---|---|
| ETH supply, 10 Nov 2026 09:36 UTC | 122,205,698 ETH |
| ETH supply at The Merge, 15 Sep 2022 | 120,520,222 ETH |
| Lowest supply since The Merge, 5 Apr 2024 | 120,063,130 ETH |
| Change since The Merge | +1,685,476 ETH (+1.40%, about 0.34% a year) |
| Burned since The Merge | 2,020,257 ETH |
| Burned in the 30 days to 10 Nov 2026 | 2,724 ETH (6.23 million dollars) |
| Net supply change, same 30 days | +86,758 ETH (about 0.86% annualised) |
| Burned in the last 24 hours | 33.27 ETH (70,674 dollars) |
The arithmetic is plain. Over those 30 days new issuance was about 89,500 ETH (the net change plus the burn, our calculation), so the burn offset about 3 percent of it. Supply did fall after The Merge while fees were still high, down to a low of 120.06 million ETH on 5 April 2024 in the same data, and it has risen ever since as fees collapsed: it passed 121.3 million on 1 January 2026. As of 10 November 2026, ether is not deflationary: its supply grows a little under 1 percent a year on the latest 30-day pace. Whether the burn will matter again depends on demand for layer-1 block space, which is exactly what the low fees of 2026 show to be weak.
Blobs: a second fee market for rollups
Since the Dencun upgrade of 13 March 2024, Ethereum has two meters. EIP-4844 added “blobs”: chunks of 128 KiB that rollups use to post their transaction data, kept by nodes for about 18 days and never readable by contracts. Blobs have their own gas (131,072 blob gas per blob), their own target and maximum per block, and their own base fee, which rises and falls with blob demand by an exponential rule rather than the 12.5 percent step. Like the execution base fee, the blob fee is burned.
The number of blobs per block has grown in steps, from 3 target and 6 maximum at Dencun to 14 and 21 since the second “blob parameter only” fork of 7 January 2026; our comparison of layer 1 and layer 2 has the full table. Under the original rule, the authors of the fix explain, the blob price could keep falling until it settled at its floor of 1 wei whenever blob demand stayed below target, so rollups paid almost nothing while asking every node to store their data, and the price stopped telling anyone how busy blobs were. Fusaka fixed that with EIP-7918, which ties a reserve price to the execution base fee: in its words, “The blob base fee will settle at a level where a blob costs at least BLOB_BASE_COST execution gas,” with that constant set at 8,192. Since 8,192 is one sixteenth of a blob’s 131,072 units, the blob price in practice cannot stay far below one sixteenth of the execution base fee.
On the morning of 10 November 2026 the blob base fee moved between 3.5 and 4 million wei per unit of blob gas, about 0.0035 to 0.004 gwei, while the execution base fee was 0.061 to 0.073 gwei. One whole blob therefore cost about 131,072 x 3.6 million wei, or roughly 0.00047 ETH, about 1 dollar at that morning’s price of 2,097.46 dollars per ETH (Etherscan). A rollup spreads that dollar over the hundreds or thousands of transactions it packs into the blob, which is why fees on layer 2 networks are counted in fractions of a cent.
Why Ethereum fees were high, and why they are low now
Three eras in Etherscan’s numbers
The table below uses Etherscan’s daily average gas price and that day’s ETH price, both from its public charts. The dollar column is our multiplication for a 21,000-gas transfer: an illustration of the average day, not what any single user paid.
| Date | Etherscan daily average gas price | 21,000-gas transfer in ETH | ETH price that day (Etherscan) | Transfer in dollars | Context |
|---|---|---|---|---|---|
| 7 Dec 2017 | 59.66 gwei | 0.001253 | $421.15 | $0.53 | CryptoKitties congestion, launched 28 Nov 2017 |
| 12 May 2021 | 281.83 gwei | 0.005918 | $3,811.29 | $22.56 | DeFi and NFT boom, before London |
| 9 Nov 2021 | 181.38 gwei | 0.003809 | $4,732.07 | $18.02 | Week of ETH’s 2021 price peak |
| 1 May 2022 | 474.57 gwei | 0.009966 | $2,826.04 | $28.16 | Otherside land sale; 2022 daily high |
| 14 Mar 2024 | 56.89 gwei | 0.001195 | $3,880.80 | $4.64 | Day after Dencun |
| 9 Nov 2026 | 0.739 gwei | 0.0000155 | $2,132.37 | $0.033 | Layer 2s on blobs, 60M gas limit |
Read the gas price column alone and the shape is clear. Yearly averages of Etherscan’s daily figures (our calculation) were 105 gwei in 2021, 49 in 2022, 34 in 2023, 20 in 2024, 3.5 in 2025 and 0.9 gwei in 2026 up to 9 November. Fees spiked when a small, fixed supply of block space met a wave of demand: an NFT game in 2017, DeFi summer in 2020, the NFT and DeFi peak of 2021, a land sale in 2022. Before London the pricing rule turned each rush into a blind auction. EIP-1559 did not make block space cheaper; it made the price predictable and burned it. The highest daily average of 2022, on 1 May, came nine months after London went live.
What explains the low fees of 2025 and 2026
The causes are on the record. Demand for layer-1 space moved to rollups, which since 2024 post their data in blobs and settle in bulk; Ethereum mainnet still processed 2,001,721 transactions on 9 November 2026 according to Etherscan, but most users’ everyday activity happens on layer 2 networks. Supply grew: validators doubled the block gas limit from 30 to 60 million between February and November 2025. And the base fee only rises when blocks run above half full, so a network with spare room lets the price drift towards zero.
Low does not mean fixed. On 29 October 2026 Etherscan’s daily average reached 8.17 gwei, the highest of the year and about eleven times the 9 November figure. The base fee follows demand within minutes, as the ten-block example shows. A rush of minting, liquidations or panic selling can still make a few hours expensive.
How to read the fee in your wallet and on Etherscan
Wallets hide most of this behind three buttons, often called low, market and aggressive. Behind them, the steps are the same.
- Look at the estimated gas units. A plain ETH transfer shows 21,000; a token transfer about 45,000 to 60,000; a swap 100,000 to 150,000 or more. An estimate in the millions for a simple action deserves a second look at what you are signing.
- Look at the base fee the wallet assumes, in gwei. Etherscan’s gas tracker shows the current one; on 10 November 2026 at 09:42 UTC it read 0.061 gwei.
- Check the max fee. Wallets usually set it at a multiple of the current base fee so the transaction survives a few rising blocks. You will not pay it unless the base fee climbs that high.
- Check the priority fee. When blocks are not full, a tip near zero is normal. Raise it only if you need the next block on a busy day.
- Multiply. Expected fee = gas units x (base fee + tip). Worst case = gas limit x max fee, which is the amount your wallet must hold on top of what you send.
- After it is mined, open the transaction on Etherscan and compare “Transaction Fee”, “Burnt” and “Txn Savings” with your estimate, as in the example above.
Labelled example: a swap of 130,000 gas at a base fee of 0.07 gwei and a tip of 0.01 gwei costs 130,000 x 0.08 = 10,400 gwei, or 0.0000104 ETH, about 2.2 cents at an example price of 2,100 dollars per ETH. With a max fee of 0.25 gwei the wallet would reserve 32,500 gwei and refund the difference.
How to pay less gas on Ethereum
The biggest saving is to move the activity off layer 1. A swap or a transfer on a rollup shares the cost of one blob with many other users, which is why it usually costs a fraction of mainnet even in 2026. The trade-offs of doing so, bridges, withdrawal times and the security model, are covered in the layer 1 versus layer 2 lesson linked above, and a dedicated lesson on layer 2 networks comes later in the course.
On mainnet itself, the levers are modest. Do not raise the tip unless you are in a hurry: in calm conditions the base fee alone gets you into the next blocks. Avoid acting in the middle of a visible rush, such as a big mint or a market crash, when the base fee can triple in minutes. Batch what you can: one transaction that does two things pays the 21,000 intrinsic cost once. And set the max fee to a level you accept, so a sudden spike leaves your transaction waiting instead of charging you more than you planned.
Claims that a particular hour or weekday is always cheapest are common but should be checked against current data before you rely on them: in 2026 Etherscan’s monthly averages moved only between about 0.6 and 1.4 gwei, so the gains from timing are small next to the gains from using a layer 2.
When a transaction fails or gets stuck
A failed transaction is not free. If your transaction is valid but runs out of gas or hits a revert during execution, every change it made is undone, yet the gas consumed up to that point is paid and its base-fee part burned. ethereum.org notes that when the gas runs out during execution, “all the gas provided will still be consumed”. The one exception is a transaction rejected at validation, for instance with a gas limit below the 21,000 minimum: it never enters a block and costs nothing.
A stuck transaction is a different problem. If your max fee sits below the current base fee, the transaction waits in the smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool. Because every Ethereum account numbers its transactions with a nonce and processes them strictly in order, as our lesson on the account model explains, every later transaction from the same account waits behind it.
The fix lives in your wallet and costs only a new fee. To speed up, the wallet resends the same transaction with the same nonce and higher fees; only one of the two can ever be mined, so you do not pay twice. To cancel, it sends a zero-value transaction to your own address with that same nonce and higher fees, which takes the slot. Nodes running Geth only accept a replacement whose price beats the original by at least 10 percent by default, which is why wallets raise both the max fee and the tip by at least that much. MetaMask’s support page describes both buttons. Start with the oldest pending nonce. Never hand your keys or seed phrase to a website or service that offers to “unstick” transactions: no legitimate fix needs them.
Gas fees and your tax return
Gas is paid in ether, and for tax purposes that payment is not neutral. The IRS’s digital asset FAQ lists “transaction and ‘gas’ fees” among digital asset transaction costs (question 53), which can reduce the amount realised on a sale or add to the basis of a purchase; costs paid only to transfer assets, for instance between your own wallets, do not count as such costs. It also says that paying such fees with digital assets means you “have disposed of the digital assets used” and recognise a gain or loss (question 97). Each gas payment is, in principle, a tiny disposal of ETH. Our guide to crypto taxes in the United States covers the rest.
In the United Kingdom, HMRC’s Cryptoassets Manual counts fees paid to have a transaction included on the ledger among allowable costs (CRYPTO22150), and treats tokens given as a fee as disposed of at their market value (CRYPTO22280). In France, the tax administration’s doctrine (BOI-RPPM-PVBMC-30-20, paragraph 50) reduces the sale price, on proof, by the fees the seller bore for that sale, naming the fees of trading platforms and of the network members who validate transactions, and it accepts that they are paid in crypto-assets. It speaks of a taxable sale; the treatment of gas spent on DeFi operations is not settled. In Italy, the Agenzia delle Entrate’s circular 30/E of 27 October 2023 (page 50) contrasts crypto-assets with financial assets, whose cost includes commissions: for crypto-assets, the rule does not allow the costs inherent in buying and selling to be taken into account when the gain is computed. None of this is advice; keep the transaction hashes, which record every fee.
What may change next: Glamsterdam’s gas repricing
The 21,000 figure and the storage costs above are protocol choices, and the next upgrade, Glamsterdam, proposes to change several of them. It has been live on the Sepolia test network since 6 October 2026. Its rehearsal on the Hoodi test network, planned for 26 October, was postponed while developers study how EIP-8037 interacts with contracts already on mainnet, and no new date has been set, as our Glamsterdam event page records. As of 10 November 2026 there is no mainnet date, and the clients’ own configuration on mainnet shows no upgrade scheduled. Its list of scheduled changes includes EIP-2780, which would replace the flat 21,000 with a cost built from its parts: a plain ETH transfer to an existing account would stay at 21,000, but a transfer to yourself would fall to 12,000 and a zero-value transfer to a normal account to 15,000.
EIP-8038 would raise the price of state access, for example cold account access from 2,600 to 3,000 gas, while a companion proposal, EIP-8037, would make creating new state (accounts, storage slots, contracts) much more expensive. The aim is to let the block gas limit rise further without the state growing out of control: cheaper computation per block, dearer permanent storage. The catch is that older contracts which assume fixed gas amounts could break, and two rival drafts were opened on 8 November 2026, as our report on the Glamsterdam repricing dispute explains. Until a mainnet date is set, every number in this paragraph is a proposal.
The next lesson of this path, smart contracts explained with real examples, looks at the programs that consume all this gas. For fees, the dates to watch are the core developers’ calls of 12 and 19 November 2026, the first since the rival drafts appeared, though neither draft had a slot on the agenda when it was filed.
