Gas fees: why you pay what you pay, and how to pay less

Key points

Gas measures computational work, not time and not value: moving $5 and moving $500,000 in the same operation costs exactly the same.
Since 5 August 2021, every Ethereum transaction pays an algorithmic base fee that is destroyed plus a priority fee that the validator keeps.
As of August 2026 gas sits at 0.151 gwei: a Uniswap swap runs about $0.131 and a plain ETH transfer around $0.01.
The collapse traces back to Dencun (13 March 2024) and EIP-4844 blobs, which gutted Layer 2 costs and pulled activity off the base layer.
If a transaction fails you still pay: the gas burned up to the point of failure is not refunded.

Market data: Market data on this page was collected on 28 August 2026 and changes constantly. Check the linked sources before relying on it.

You open your wallet to send $40 in stablecoins, and just before you sign, a line appears reading "estimated fee: 0.0000032 ETH". You have no idea whether that is a lot or a little, why the same operation asked for five times more last week, or why a friend paid $90 for the identical thing back in 2021.

Gas is probably the worst-explained concept in Ethereum, and the analogies are to blame. People say it is "like petrol in a car", which tells you nothing you can act on. The reality is simpler and far more useful: gas is a unit of measurement for computational work, and its price is an auction market that resets every twelve seconds.

This guide takes the bill apart piece by piece, with the real numbers as of August 2026, and finishes with the only part that actually matters — what you can do to pay less.

Gas measures work, not time and not value

Every operation the Ethereum Virtual Machine can perform carries a fixed cost in gas units, written into the protocol itself. Adding two numbers costs 3 units. Writing a fresh value into permanent storage costs 20,000. A plain ETH transfer between two accounts costs exactly 21,000, always, and has done since the first block.

Two consequences follow, and both surprise almost everyone.

The amount you send is irrelevant. Sending $5 or sending $500,000 in ETH burns the same 21,000 gas units. A bank charges a percentage; a blockchain charges for computation. That is a structural difference, not a promotional rate.

Complexity, on the other hand, matters enormously. A swap on a decentralised exchange touches several contracts, reads prices, updates balances and emits events. It consumes something like fifteen times the gas of a transfer. That is why interacting with smart contracts always costs more than simply moving coins around.

The final bill is a multiplication of two independent factors: gas consumed × gas price. Code fixes the first. The market fixes the second.

What changed on 5 August 2021

Before the London upgrade, gas pricing was a blind auction. You named a price, and if you undershot, your transaction sat unconfirmed for hours. The result was a miserable user experience and fees that spiked unpredictably.

EIP-1559, live since 5 August 2021, split the fee into two parts with completely different destinations.

The base fee: algorithmic, and destroyed

The protocol computes a base fee for each block by looking at how full the previous one was. If that block ran above the target occupancy, the base fee rises; if it came in below, it falls. The adjustment is capped at 12.5% per block, which makes the price predictable in the short term: you can be reasonably confident what gas will cost sixty seconds from now.

Here is the part that matters: nobody receives that base fee. It is burned. The ETH paid as base fee leaves circulation permanently. When network activity is high and the burn outpaces new ETH issued to validators, total supply shrinks. When the network is quiet — as it has been for much of 2026 — the burn is negligible and supply grows again. This is not a guaranteed deflationary mechanism. It is a mechanism that depends entirely on usage.

The priority fee: the only part you negotiate

On top of the base fee you attach a priority fee, a tip that goes in full to the validator who includes your transaction. It is the only thing you compete on. When your wallet offers you "slow / normal / fast", it is adjusting that tip and nothing else.

You also set a max fee: the ceiling you are willing to pay per unit of gas in total. If the base fee turns out lower, the difference comes back to you. That automatic refund is the reason almost nobody wildly overpays any more.

Gas limit, gas used, and the awkward case of the failed transaction

Three distinct numbers that wallets mix together without mercy:

Gas limit: the ceiling of units you authorise the transaction to consume. Your safety cap.
Gas used: what execution actually consumed.
Gas price: base fee plus priority fee, quoted in gwei.

If the gas limit is set too low, execution stops halfway and the transaction reverts with an out-of-gas error. Chain state returns to where it was, but the gas spent up to the failure point is gone. The same thing happens when a swap reverts because the price moved beyond your slippage tolerance.

A failed transaction is not free. Validators executed your code up to the error and they charge for that work. You are only refunded the slice of the gas limit you never touched. With gas at 0.151 gwei it barely stings; during a congestion spike, a failed swap could vaporise tens of dollars and hand you nothing.

Why you pay cents in 2026

In May 2021 and through the 2022 peaks, gas moved between 50 and 200 gwei. As of August 2026 it sits at 0.151 gwei across the low and average tiers, and 0.166 gwei at the high tier, according to Etherscan's gas tracker. That is a fall of three orders of magnitude, and it was neither an accident nor simple loss of interest.

Dencun, 13 March 2024. This upgrade shipped EIP-4844, known as proto-danksharding. Until then, Layer 2 networks had to write all their data into Ethereum calldata, competing for the same expensive space as every ordinary transaction. EIP-4844 created a separate, temporary space — blobs — with its own independent fee market and automatic deletion after a few weeks. The cost of running an L2 collapsed overnight, and with it the price passed on to you.

The migration of activity. Once an L2 costs cents, users and protocols move there. That drains congestion from the base layer: the demand that used to set Ethereum's base fee has largely relocated. The L1 has become a settlement layer, and its gas market reflects that.

Fusaka and the BPOs. On 3 December 2025 the Fusaka upgrade activated PeerDAS and raised the per-block gas limit to 60 million. Two blob-parameter adjustments followed: BPO1 on 9 December 2025 (target of 10 blobs, maximum 15) and BPO2 on 7 January 2026 (target 14, maximum 21). More blob space means less pressure on blob pricing, and therefore cheaper L2s still.

The numbers, as of August 2026

Network	Typical transaction cost	Date of figure
Ethereum L1 — ETH transfer	~$0.01 (21,000 gas × 0.151 gwei)	28 Aug 2026
Ethereum L1 — Uniswap swap	~$0.131	28 Aug 2026
Ethereum L1 — general observed range	$0.04 – $0.24	28 Aug 2026
Base	~$0.02 (median cost)	24 Apr 2026
Optimism (OP Mainnet)	~$0.03	24 Apr 2026
Arbitrum One	~$0.04	24 Apr 2026
zkSync Era	~$0.05	24 Apr 2026
Bitcoin — typical ~140 vB transaction	140–280 sats ≈ $0.11–$0.22	28 Aug 2026

Reference prices used: ETH at roughly $2,505 and BTC at roughly $79,560 on 28 August 2026. Every one of these values is volatile by definition. Check them yourself before making any decision on them.

Worked example: a $500 swap on L1 and on Base

You want to swap $500 of stablecoins into ETH. Gas costs are quoted in dollars throughout, and as you will see, the gas is so small that exchange rates change nothing about the conclusion.

Step 1: how much gas a swap actually burns. You do not have to guess this — you can derive it. If the swap costs $0.131 with ETH at $2,505, that is 0.131 / 2,505 = 0.0000523 ETH. Divide by the gas price (0.151 gwei = 0.000000000151 ETH per unit) and you get around 346,000 gas units. That lines up with a typical router path: sixteen times what a plain transfer costs.

Step 2: the breakdown on Ethereum L1.

Component	Calculation	Cost
Base fee (burned)	346,000 gas × 0.151 gwei	$0.131
Priority fee to validator	346,000 gas × ~0.015 gwei (gap between low and high tiers)	$0.013
Total gas		~$0.144
Pool fee	Depends on the specific pool's fee tier	Not estimable without knowing it
Slippage	Depends on liquidity depth	Not estimable without knowing it

Out of your $500, gas takes roughly $0.14 — about 0.03% of the amount. What actually determines how much you end up with is the pool fee and the slippage, not the network.

Step 3: the same swap on Base. The median transaction cost on Base was $0.02 as of 24 April 2026. A swap consumes more than the median transaction, but you are still talking about fractions of a cent to a few cents. In absolute terms the gap versus L1 is around ten cents.

Step 4: the trap in this example. If your $500 sits on Ethereum L1 and you want to trade on Base, you have to bridge it. The bridge costs one L1 transaction (on the order of $0.13, plus whatever the bridge itself charges) and exposes you to contract risk that the L1 does not carry. Bridging to save ten cents is absurd. Bridging because you are about to make thirty trades makes perfect sense.

Step 5: the 2021 contrast. Those same 346,000 gas units, at 100 gwei, would have cost 0.0346 ETH. At the August 2026 ETH price of $2,505, that is $86.60 for a single swap. The change is not cosmetic: whole categories of application were simply not viable in 2021.

Bitcoin charges on a different basis

Bitcoin has no virtual machine and no gas, which is one of the structural differences between the two networks. Its fee depends on transaction size in virtual bytes (vB), not on the amount sent and not on computation.

Size depends on how many inputs (UTXOs) your transaction spends. If your bitcoin is scattered across forty small receipts and you want to spend the lot, the transaction will be large and expensive. If it sits in two big UTXOs, it will be small and cheap. This is why experienced users consolidate UTXOs when the mempool is quiet: they merge many inputs into a single output while fees are low, so that future transactions are cheaper.

As of August 2026 the mempool is very empty and fees run at 1–2 sat/vB. A typical transaction of roughly 140 vB costs between 140 and 280 sats, which is $0.11 to $0.22 with BTC at $79,560. You can check the live state at mempool.space.

Bitcoin also has no nonce-replacement mechanism like Ethereum's. Its equivalent is RBF (Replace-By-Fee): you flag the transaction as replaceable when you broadcast it, then rebroadcast later with a higher fee. If you did not flag it, the fallback is CPFP — spending the pending output in a new, high-fee transaction so the miner has an incentive to confirm both together.

What to do to pay less

Check the gas price on Etherscan's tracker before you sign. If you are deploying a contract or running a 500,000-gas operation, waiting for a trough genuinely changes the number.
If you plan more than two or three operations, work on an L2 and bridge once. The saving lives in transaction volume, not in any single trade.
Review the gas limit your wallet proposes when you interact with an unusual contract. A limit set too tight causes an out-of-gas revert that you pay for and get nothing from.
On Bitcoin, consolidate UTXOs while the mempool sits at 1–2 sat/vB. It is an investment: you spend a cheap fee today so every future transaction is smaller.
Learn how your wallet's speed up works. It is a transaction with the same nonce and a bigger tip, not a new operation.
Reuse approvals where the protocol allows it, rather than approving and revoking on every single trade.

Do not max out the priority fee "just in case". With an algorithmic base fee and automatic max-fee refunds, you buy almost nothing with it.
Do not bridge to an L2 to save a few cents on one operation. The bridge costs more than the saving and adds contract risk.
Do not assume the low costs of 2026 are permanent. The base fee is a market, and a strong demand event multiplies it within a handful of blocks.

Where things stand in late 2026

Ethereum's roadmap keeps pushing in the same direction. Glamsterdam, bringing ePBS and Block-Level Access Lists, targets 4 November 2026 and is running on public testnet. Hegotá aims for May 2027 with FOCIL and progress on statelessness. Neither promises lower L1 fees specifically, but both are about efficiency and about separating roles within the block more cleanly.

The practical reading is that Ethereum has accepted its role: the base layer settles and guarantees, the L2s execute. Whether that convinces you depends on how much weight you give to the risks each L2 adds, which are not trivial. What is no longer in dispute is the arithmetic: as of August 2026, the cost of using Ethereum has stopped being the barrier it once was.

Frequently asked questions

Why was I charged gas when my transaction failed?

Because validators executed your code up to the point where it broke, and that work has to be paid for. When a transaction reverts, the chain state rolls back but the gas already consumed does not come back. You are only refunded the portion of the gas limit you never reached. That is why a badly configured approval or a swap with too little slippage tolerance costs you money and gives you nothing.

What is a gwei, and why not just quote prices in ETH?

A gwei is one billionth of an ETH (0.000000001 ETH). It exists because gas prices in ETH are absurdly small numbers that nobody can read at a glance. As of August 2026 gas sits at 0.151 gwei, which is 0.000000000151 ETH per unit of gas. Multiply that by the 21,000 units a simple transfer needs and you land at roughly one US cent.

Can I cancel a stuck transaction?

You do not cancel it, you replace it. Every transaction carries a nonce, a sequential counter for your account. You broadcast a second transaction with the same nonce and a higher priority fee; validators pick whichever pays more, and the original becomes invalid. Send a 0 ETH transaction to yourself with that nonce and the original disappears entirely. Wallets label this speed up and cancel.

Is gas cheaper overnight?

Historically yes, and the logic still holds: the base fee rises and falls with how full the previous block was, and blocks fill up when the US and Europe are awake. What changed in 2026 is the magnitude. With gas at 0.151 gwei, waiting six hours saves you fractions of a cent. It is worth timing a contract deployment, not a transfer.

Sources and references

Important notice. InfoCrypto is an educational site. Nothing here is financial, tax or investment advice. Crypto assets are high risk: you can lose all the capital you invest. Always verify data against the original sources before making any decision, and consult a licensed professional if you need to.