Ethereum Gas Fees, Explained: What You Pay and Why
Ethereum Gas Fees, Explained: What You Pay and Why
Analysis by CryptosEyes Research | Updated October 1, 2026
Short Answer
Every Ethereum transaction has two independent cost levers: how much computation it requires (measured in gas units) and what you pay per unit (measured in gwei). Your fee is gas used multiplied by gas price, and the gas price itself has two parts: a protocol-set base fee that gets burned, and a priority fee, the tip you pay the validator.
Typical mainnet conditions in 2026 look nothing like 2021. Multiple data sources put the average base fee somewhere between 0.1 and 1 gwei for most of this year, down from a 2021 average of roughly 80 to 180 gwei. A simple ETH transfer, which always costs 21,000 gas units, now runs well under five cents. A token swap that cost around $86 at peak congestion in 2021 can cost under twenty cents when the network is quiet. The structural reason is the rollup-centric roadmap: about 95 percent of Ethereum transactions now happen on layer 2 networks, and blob upgrades (EIP-4844, Pectra, Fusaka) made it dramatically cheaper for those networks to post their data to mainnet.
Two Things Decide Every Fee: Units and Price
Gas is a unit that measures computational work, the same way a kilowatt-hour measures electricity. A validator's computer has to execute your transaction, and gas meters exactly how much work that execution requires. The price you pay per unit of gas floats with demand. Your total fee is:
Fee = gas units used x (base fee + priority fee), with the result priced in gwei
The units side is mostly out of your control and mostly predictable. A plain ETH transfer always costs exactly 21,000 gas, set by the protocol. Contract interactions cost more because they do more work. Published figures across 2026 reference material are consistent:
| Action | Typical gas used | What makes it cost that much |
|---|---|---|
| ETH transfer | 21,000 | Fixed by protocol; moves a balance |
| ERC-20 token transfer | ~45,000 to 65,000 | Updates the token contract's balance records |
| ERC-20 approval | ~46,000 | Records a spending allowance for another contract |
| ERC-721 NFT mint | ~150,000 | Writes new token data and ownership records |
| Uniswap V3-style swap | ~150,000 to 200,000 | Multiple pool state reads and writes, routing logic |
| Simple contract deployment | ~500,000 | Writes full bytecode to chain |
| Complex DeFi contract deployment | ~3,000,000 | Large bytecode plus initialization storage |
These numbers come from engineering references and educational sources cited at the end; treat them as order-of-magnitude guides, not quotes for your specific transaction.
Two details matter more than beginners expect. First, the gas limit you set in your wallet is a ceiling, not a price tag: if you set 200,000 but use 65,000, the rest is never charged. Set the limit too low and the transaction fails, while the gas consumed up to the failure point is still charged. Second, you pay whether the transaction succeeds or fails. A reverted contract call changes nothing on chain, but validators still did the work up to the revert, so you pay for it. Always estimate before firing an unfamiliar transaction.
gwei: The Unit That Makes the Numbers Readable
Fees are quoted in gwei because quoting in ETH would be unreadably small. One gwei is 0.000000001 ETH, one-billionth of an ether. At one gwei and a 21,000-gas transfer, the cost is 21,000 x 0.000000001 = 0.000021 ETH. Convert to dollars with the current ETH price and you get the figure that actually matters.
A hypothetical example: suppose the base fee sits at 0.5 gwei, you add no tip, and ETH is at $1,878 (a mid-2026 figure referenced by CoinPaprika). A simple transfer: 21,000 x 0.5 = 10,500 gwei = 0.0000105 ETH, about $0.02. A 180,000-gas swap under the same conditions: 180,000 x 0.5 = 90,000 gwei = 0.00009 ETH, about $0.17. Both figures are snapshots; run this arithmetic with live tracker numbers rather than memorizing the outputs.
EIP-1559: How the Base Fee, the Tip, and the Burn Work
Before August 2021, Ethereum ran a blind first-price auction. You guessed a gas price, submitted, and hoped. Bid too low and your transaction sat in the mempool for hours; bid too high and you overpaid with no refund. EIP-1559 replaced that auction with a structured fee model that still runs today.
Each block has a base fee, set by the protocol, that must be paid for any transaction to be included. The base fee adjusts every block based on how full the previous block was:
The mechanism self-regulates toward blocks that are half full. Crucially, the base fee is burned: it is removed from ETH supply rather than paid to anyone. This is the burn mechanism you hear about in ETH supply discussions. Because fees rise with network usage and that fee revenue is destroyed, periods of heavy usage can make ETH supply deflationary, while quiet periods keep issuance positive.
On top of the base fee, you add a priority fee, the tip that goes to the validator. Tips are the only part of the fee you control directly. During quiet conditions, a tip of 0.01 to 0.1 gwei is enough; when blocks are full, validators pick transactions with higher tips first, so a larger tip buys faster inclusion.
Modern wallets submit Type 2 (EIP-1559) transactions with two fields: maxFeePerGas, the most you will pay per unit, and maxPriorityFeePerGas, the most you will tip. You only ever pay the actual base fee plus your tip, so a generous maxFeePerGas is headroom against spikes, not a commitment. Legacy Type 0 transactions with a single flat gasPrice still work, but almost everything today is Type 2.
The 12.5 percent cap is per block, and blocks land roughly every 12 seconds, so sustained demand compounds fast. Illustrative math: 1.125 to the 10th power is about 3.2, so ten consecutive full blocks can roughly triple the base fee in about two minutes. That is why fees feel calm, then violent: the cap keeps any single block orderly, but compounding across blocks gives demand shocks room to bite.
Why Fees Spike: The Four Engines of Congestion
Spikes are not random. They come from identifiable demand shocks colliding with fixed blockspace.
1. NFT mints and token launches. A hyped mint sends thousands of wallets racing to be included before the supply sells out. Historical statistics place peak congestion gas above 200 gwei during major NFT drops, when everyone bids against everyone else for the same 12 seconds of blockspace.
2. DeFi liquidations. When markets drop hard, collateralized positions get liquidated, and liquidation bots compete to execute those liquidations first. Our companion piece on crypto liquidations covers the mechanics in full, but the fee angle is simple: liquidation races are winner-takes-all, so participants bid tips aggressively, and the resulting base-fee compounding raises prices for every ordinary user caught in the same blocks.
3. L2 batch posting. Rollups must post compressed transaction data to mainnet on a schedule, and when blob demand pushes against the blob supply target, the blob base fee rises through its own adjustment mechanism. That cost flows back into L2 transaction fees. In other words, one of the things that makes L2 fees occasionally hiccup is the settlement bill those networks owe Ethereum.
4. MEV bot wars. When an arbitrage or liquidation opportunity appears, competing bots bid against each other in the same block. Fee statistics sources describe these wars as capable of driving gas 10 to 20 times above baseline for short stretches. As an ordinary user you cannot win that race; the correct response is to wait it out.
General mania sits on top of all four. In 2021, sustained speculative demand held average gas between 80 and 180 gwei, with peaks above 1,000 gwei. Those peaks are history, not a forecast, but the mechanism that produced them is still in the protocol.
2021 vs 2026: What the Numbers Look Like Now
The scale of the change is worth a direct comparison. Figures below are drawn from the fee-history and fee-statistics sources cited at the end; label the column, not your memory, as the source of truth.
| Metric | Peak congestion (2021-2024) | Typical 2026 |
|---|---|---|
| Average gas price | 72-220 gwei | 0.1-1 gwei most days |
| Extreme peak gas | 1,000+ gwei | ~2 gwei (year-to-date high per one tracker) |
| Record low | n/a | 0.067 gwei, a November 2025 anomaly |
| Simple ETH transfer | $20-$50 | ~$0.02 |
| Token swap | ~$86 | Under $0.50 when quiet |
| NFT mint | ~$145 | Under $1 when quiet |
| Daily network fee revenue | ~$23 million | ~$7.5 million |
September 2026 shows day-to-day variance: September 6 averaged about 0.68 gwei while September 15 averaged about 1.13 gwei, per CoinLaw's statistics, so the same transaction cost materially more on the busier day. One tracker reported the most recent daily average near 0.90 gwei, about 28 percent below the same point a year earlier. Fees are low and trending lower, but they still breathe.
Blobs: What EIP-4844 Changed for L2 Fees
To understand 2026 fees, you need blobs. Before March 2024, rollups posted their transaction data to mainnet as calldata, competing for the same expensive execution blockspace as everyone else. EIP-4844, activated in the Dencun upgrade, created blobs: large, cheap, temporary data packages designed specifically for rollup data, priced in their own separate fee market.
The practical details, from engineering references:
The effect on L2 economics was the point of the whole roadmap. Blob posting cut L2 data costs roughly 90 to 99 percent versus calldata, and analyses of the Fusaka-era capacity suggest a further 60 to 95 percent reduction in average L2 data costs depending on demand. That is why an L2 transaction that once cost dollars now costs cents or fractions of a cent, and it is why the mainnet base fee fell: with roughly 95 percent of transactions executing on rollups, far less demand competes for L1 execution blockspace.
The Anatomy of an L2 Fee
A transaction on a rollup has two cost components, and distinguishing them explains most L2 fee surprises:
Early-2026 engineering data on a swap on Base put the L2 execution share near $0.0003 and the L1 blob share near $0.0027, for a total near $0.003. Between August 21 and September 20, 2026, Base's settlement cost to Ethereum averaged about $0.000035 per user operation, according to CoinLaw's fee statistics. Simple transfers on Base run around $0.001 or less, and typical DEX swaps land in the $0.001 to $0.01 band. These are early-to-mid-2026 figures; the structure is stable, the cents move.
Two takeaways. First, L2 fees are dominated by the data bill, not execution. When your L2 fee is higher than usual, it is usually the blob market passing through a cost. Second, fee competition between rollups is largely competition in how efficiently they pack transactions into blobs. Our coverage of the layer 2 wars and the top layer 2 solutions covers which networks are winning that efficiency race and why it matters for fees.
How to Read a Gas Tracker Without Fooling Yourself
A gas tracker is a dashboard for a market you cannot see directly, the mempool. What each field means:
What trackers cannot tell you: whether a demand shock is about to arrive. A tracker reads the present mempool; it cannot know that a hyped NFT mint starts in ten minutes or that a liquidation cascade is brewing in the next block. Treat tracker predictions as measurements with a minutes-long half-life, not as forecasts.
Paying Less: Tactics That Actually Work
1. Move the activity to a layer 2. This is the structural answer and it dwarfs everything else. If the application you use exists on a rollup, transacting there instead of mainnet cuts fees by roughly an order of magnitude or more in typical conditions.
2. Time your mainnet transactions. Fees follow human activity: weekends and off-peak hours are historically cheaper than weekday business hours. Check the tracker's daily pattern before sending anything non-urgent. Remember the September 2026 data: 0.68 gwei on a quiet day versus 1.13 on a busy one is a meaningful difference on a large transaction.
3. Batch and plan your approvals. An ERC-20 approval costs about 46,000 gas and a token transfer another 45,000 to 65,000. If you approve a DEX and then trade, you pay both. Some wallets and smart-account flows (EIP-7702 transaction types, which arrived with Pectra) can bundle actions into fewer transactions. Where bundling is not available, avoid redundant approvals: approving the exact amount rather than unlimited, and not re-approving a contract you already approved.
4. Set headroom, not overpayment. A maxFeePerGas well above the current base fee protects you if fees rise while your transaction waits; you only pay the actual base fee, not the maximum. What wastes money is a maxPriorityFeePerGas far above what inclusion needs. During quiet periods, the 0.01 to 0.1 gwei tip band cited by fee references is plenty.
5. Respect failed transactions. A failed or reverted transaction still costs gas, sometimes the entire limit. Test unfamiliar contract interactions with small amounts, read the error before retrying, and do not spam the same failing call: each attempt bills you again.
6. Do not revoke approvals gratuitously. Revoking a stale approval is itself a transaction with a fee. If the approval is to a reputable contract you might use again, revocation can cost more than the risk it removes. Treat revocation as maintenance for contracts you no longer trust, not as a reflex.
The Real Limits of Fee Prediction
Fee prediction is worse than most people think, and the reasons are structural.
First, every public estimator is reactive. It looks at the mempool and recent blocks, then extrapolates. It has no window into demand that has not arrived yet: the NFT mint scheduled for next hour, the liquidation cascade triggered by a price move in the next block, the MEV war that starts without warning. Estimates with a minutes-long horizon are the ceiling, not the floor, of what prediction can do.
Second, the compounding math cuts both ways. The same 12.5 percent per-block adjustment that normally keeps fees boring can triple the base fee in about ten full blocks. An estimate made during calm conditions can be stale within two minutes of a real demand shock.
Third, you are predicting two markets, not one. Mainnet transactions face the execution base fee; L2 fees pass through the blob base fee. A quiet execution market with a hot blob market still raises your L2 costs, and most consumer-facing trackers only show the first clearly.
Fourth, inclusion time is a probability, not a schedule. Low, average, and fast presets describe your odds of landing in the next few blocks, not a queue position. Validators choose by tip when blocks are full, and a sudden bid war can push a "fast" transaction down the ordering.
A future upgrade, Glamsterdam, is in the pipeline and is expected to push the fee floor lower again, according to 2026 fee analysis. That is a structural direction, not a timetable for your next transaction. The practical posture stays the same: check the tracker, pick the right network, set sane headroom, and never send a transaction you cannot afford to have fail.
Frequently Asked Questions
Why did my transaction fail but I still got charged?
Because validators did the work. Fees pay for computation, not results. Your transaction consumed gas executing up to the point it reverted, and the protocol charges for that computation whether the state change committed or not. This is also why setting an accurate gas limit matters: a limit too low guarantees a failure that you still pay for.
What is the difference between gas limit and gas used?
The gas limit is the maximum computation you authorize; gas used is what the transaction actually consumed. You are charged for used, not for limit. Set the limit high enough that the transaction cannot run out of gas mid-execution, and the unused portion costs you nothing.
Should I always pick the "fast" preset in my wallet?
No. The fast preset bids above the current base fee plus a generous tip, which is correct when you need the next block and wasteful otherwise. For non-urgent transactions, the standard preset, or a custom tip near 0.01 to 0.1 gwei in quiet conditions, gets you confirmed a few blocks later for less.
Why is my L2 transaction fee higher than usual?
Because of the L1 data component. L2 execution itself is cheap and stable; the variable part is your share of the blob posting cost. When blob demand pushes against the supply target, the blob base fee rises and rollups pass it through. If a tracker shows a calm mainnet base fee but your L2 fee is elevated, the blob market is the usual suspect.
Do fees differ between wallets?
The protocol charges the same for the same transaction, but wallets set different defaults: networks, tip presets, maxFeePerGas headroom. Two wallets executing the identical swap can produce meaningfully different fees purely from presets. Check the network and fee fields before confirming, every time.
Can gas fees go back to 2021 levels?
The mechanism that produced 2021 fees is still in the protocol: sustained demand against fixed blockspace compounds the base fee 12.5 percent per block. What changed is the demand distribution: roughly 95 percent of transactions now execute on L2s, and blob capacity keeps growing. A return to 100+ gwei averages would require mainnet execution demand to surge far beyond current patterns. Possible in theory, and no one can rule it out, but the structural direction since 2024 has been capacity expanding faster than mainnet demand.
Related Reading
Sources
Gas prices, ETH prices, and upgrade schedules move. The figures above reflect sources current to September and October 2026. Recheck a live tracker before acting on any specific number.
CryptosEyes publishes general educational research, not investment, legal, or tax advice. Crypto assets and on-chain activity carry risk, including total loss. Protocol parameters and network conditions can change after publication.
Source & Review Basis
This article is reviewed against the source types below. Source links are provided to help readers verify primary documents, market context, and methodology independently.