A layer 2 is a separate blockchain that runs transactions itself and posts the data back to Ethereum. That one design choice explains the whole fee gap. An Ethereum layer 2 does not make every node on the most expensive chain replay your swap. It does the work off to the side and pays Ethereum for settlement and data only.

The numbers are not subtle. Per Ethereum.org’s L2 directory snapshot for 1–6 October 2026, Base displayed an average fee of $0.001–0.002 per transaction, while a plain ERC-20 transfer on Ethereum layer 1 typically cost $2–10. Choosing an Ethereum layer 2 in 2026 is therefore less about saving money, which is settled, and more about which liquidity, bridge route and governance model you want to live with.

Some links on this page are affiliate links. If you sign up through them we may earn a commission at no extra cost to you. Offers, bonus amounts and expiry dates are verified daily but can change without notice — always read the official terms before you invest. Crypto assets are volatile and this page is not financial advice.

## What an Ethereum layer 2 actually is

### Rollups post data, not execution

A rollup runs its own chain with its own blocks and gas market. It bundles hundreds of transactions, compresses them, and publishes the result to Ethereum in a batch. Ethereum stores that data and settles disputes. It does not re-run your trade, which is why it can charge a fraction of a mainnet fee and still be honest about the accounting.

EIP-4844, shipped in 2024, changed the economics again. It added [blobs](https://ethereum.org/en/roadmap/danksharding/), a cheaper data channel with a shorter retention window that rollups use instead of expensive calldata. Data cost is the single largest line item in an L2 fee, so blobs are the reason the figures below are measured in fractions of a cent rather than in dimes. If you want the base-layer version of this, start with [what a blockchain is](/what-is-blockchain/) and work forward.

### Optimistic versus validity rollups

An optimistic rollup assumes each posted batch is correct and gives anyone a window to prove otherwise. A validity rollup, usually called a ZK rollup, attaches a cryptographic proof to each batch so the new state is verified before it is accepted. Optimistic systems are cheaper to run for general-purpose EVM code. Validity systems settle faster but are harder to build, and they come with different tooling, different wallet support and occasionally a different gas token.

### Why the withdrawal wait exists

Getting money into an L2 is fast, because the chain can credit a deposit as soon as it sees the Ethereum transaction. Getting it out is the slow direction. On an optimistic rollup your withdrawal sits inside the dispute window until nobody can challenge the batch that contains it. That is the entire 7 days: not congestion, just the challenge period doing its job. Third-party bridges sell faster exits by fronting the money and carrying the risk themselves, which is a different trust model, not a faster chain.

## Fees first: what an Ethereum layer 2 transaction costs

The table reproduces displayed average fees from [Ethereum.org’s L2 directory](https://ethereum.org/en/layer-2/) snapshot for 1–6 October 2026 and market-share figures from the September–October 2026 snapshot. These are dated averages, not live quotes. A quiet Sunday and a launch-day mint on the same network can differ by an order of magnitude, and the directory number is a rounded headline rather than a price you will get. The mechanics are covered in [crypto gas fees explained](/crypto-gas-fees-explained/).

| Network | Rollup type | Displayed avg fee | Market share | Withdrawal to Ethereum | Gas token |
|—|—|—|—|—|—|
| Base | Optimistic, OP Stack | $0.001–0.002 | $14.6B | 7 days | ETH |
| Arbitrum One | Optimistic | $0.004–0.005 | $12.3B | ~6.4 days, 45,818 blocks | ETH |
| OP Mainnet | Optimistic with fault proofs | $0.00 rounded | $1.65B | 7 days | ETH |
| Scroll | ZK validity | $0.002–0.003 | $45.7M | Faster than optimistic | ETH |
| Linea | ZK validity | ~$0.024 | $352M | Faster than optimistic | ETH |
| Starknet | STARK validity | $0.005–0.008 | $389M | No 7-day dispute window | STRK |
| ZKsync Era | ZK validity | Not listed in snapshot | $227M | 3-hour minimum | ETH |

Two rows need a caveat. OP Mainnet’s fee rounds to $0.00 at the precision the directory displays. It is not free, just below a cent. ZKsync Era has no fee figure in the same snapshot, so we left it blank rather than invent one. Its distinguishing number is the 3-hour minimum exit, which belongs in the withdrawal column.

Market share is the more useful column. Base and Arbitrum One hold about $26.9B between them out of roughly $29.6B across the seven networks. The long tail is small: Scroll’s $45.7M is a third of a percent of Base’s $14.6B. Scale sets depth, and depth decides whether a large swap moves the price against you.

Fee differences among the top three are smaller than the measurement noise. Base and Arbitrum One differ by three-tenths of a cent. One Ethereum withdrawal fee, a flat exchange charge, is worth more than a thousand transactions at that gap. Choosing on the fee column means optimizing the wrong variable.

## Base vs Arbitrum vs OP Mainnet: the three that matter

### Base

Base is the largest Ethereum layer 2 by value, with $14.6B in the September–October 2026 snapshot, and the cheapest of the three at a displayed $0.001–0.002. It runs on the OP Stack, sharing its codebase with OP Mainnet. The feature that changes behaviour is Coinbase: you can move dollars from a Coinbase account straight onto Base without touching a bridge, which is the shortest path from fiat to an L2 that exists today. The exit to Ethereum takes 7 days. The trade-off is concentration. One company operates the sequencer and sits on the upgrade path, and no token vote can overrule it.

### Arbitrum One

Arbitrum One is second at $12.3B and the deepest venue of the three for DeFi liquidity. If you are moving size into lending markets, perps or concentrated liquidity positions, this is generally where the book is thickest and slippage smallest. Its dispute window is 45,818 blocks, roughly 6.4 days, so its exit is marginally faster than the 7-day standard. Fees sit at $0.004–0.005, still under a cent. Its structure is committee-heavy: a security council holds emergency upgrade powers while the DAO controls the treasury and roadmap.

### OP Mainnet

OP Mainnet is the smallest of the three at $1.65B, and its case is least about money. Its displayed fee rounds to $0.00, so on cost it competes with Base, and its 7-day exit is the standard optimistic wait. What it offers instead is the reference implementation of the OP Stack, fault proofs in its directory entry, 2-second blocks, and governance built around the Optimism Collective and the OP token, including the retroactive public-goods funding model other OP Stack chains plug into. If you want the chain whose direction is set by a token vote rather than a company, this is it. You pay in liquidity: $1.65B is roughly a ninth of Base.

### The honest trade-off

Fees are near-identical across all three once you allow for display rounding, so the fee column will not decide this for you. Pick on liquidity if you trade size, because $12.3B against $1.65B is the difference between a market and a queue. Pick on bridge route if you start from fiat or an exchange balance, because the cheapest chain is the one you reach without paying two sets of fees. Pick on governance if you care who can change the contract under you. Those questions produce different answers for different people, which is why no single best-L2 claim survives contact with a real portfolio. If your starting point is an exchange account, the [best crypto exchanges](/best-crypto-exchanges/) list determines which networks you can reach directly.

## The ZK rollups: faster exits, different tooling

Scroll is the cheapest of the ZK group at $0.002–0.003 and the smallest by far at $45.7M. It exits to Ethereum faster than the optimistic rollups. It is a reasonable place for a small position and an uncomfortable place for a large one, because thin liquidity turns a routine exit into a price event.

Linea is the most expensive network in this table at about $0.024 per transaction, roughly ten times Base, on $352M of market share. You are paying for a validity proof and a faster exit. Whether that premium is worth it depends on how often you move in and out; held for months, the fee difference is irrelevant and the liquidity difference is not.

Starknet holds $389M and charges $0.005–0.008, but it pays gas in STRK rather than ETH. That changes onboarding: you need the token before you can transact, which means an extra swap and a second asset to track. In exchange you get a validity rollup with no 7-day dispute window.

ZKsync Era is the smallest of the four at $227M and enforces a 3-hour minimum withdrawal. That is the fastest documented exit here and a genuine advantage if your pattern is bridge in, swap, bridge out. It is also the clearest example of the trade: the network with the best exit latency holds the least value.

The pattern is consistent across all four. ZK buys latency on the way out and charges for it in depth, tooling maturity and sometimes gas-token complexity. For a single swap with a fast return that is a fair trade. For size parked over a season it is not, because exits need a counterparty.

## What you are trusting

Every network in the table runs a single sequencer. One node orders transactions, and it can delay, reorder or refuse them. It generally cannot forge a withdrawal, but it can degrade your experience for as long as it likes. That is a real trust assumption, not a technical footnote.

Upgrade keys are the second assumption. The contracts holding bridged assets are upgradeable, and a multisig can change them. Ask who signs, how many signatures are needed, and whether the timelock is long enough to exit before a change lands. A 3-of-5 multisig with no delay is a very different proposition from a 9-of-12 council with a week-long timelock.

Security councils exist to handle exactly that: emergency upgrades when something is exploited, and a brake on governance capture. Base, Arbitrum One and OP Mainnet all run a version of this, with different powers and different transparency. Fault proofs are the fourth piece. OP Mainnet’s directory entry lists them, and a fault proof is only as good as the window in which someone can submit one.

The practical consequence is boring and important. Keep a working balance on Ethereum mainnet or an exchange for anything you might need this week. Do not park rent money on a chain whose exit takes 6.4 days, however cheap the swap was.

## How to choose an Ethereum layer 2

Four rules cover most cases.

1. **Start where your money already is.** If your balance sits on Coinbase, Base is one withdrawal away and costs a flat fee. If it sits on a DeFi-native wallet, Arbitrum One’s depth is usually the better landing spot. Crossing two bridges to save a tenth of a cent is a losing trade.
2. **Choose liquidity over the fee decimal.** The fee column is noise between the top three and meaningful only against Linea’s $0.024. Depth and slippage are where real money is lost or saved, and the ranking there is Arbitrum One first.
3. **Check the exit before the entry.** Confirm the withdrawal path and duration before you deposit. If a 7-day wait breaks your plan, use a ZK rollup with a 3-hour exit, or keep the position where you can leave quickly.
4. **Match the asset to the network.** Native USDC and bridged USDC are not the same token, and a bridged version is an IOU from the bridge rather than the underlying asset. Verify what you are actually holding, and how it converts back.

The bridge route is where most losses happen. Use the canonical bridge or a reviewed third-party bridge, never an unfamiliar one you found in a chat, and verify that both ends support the asset before you send anything. A bridge that lists the token on the destination but not the source will happily take your deposit and leave you stuck. If your goal is simply the cheapest possible transfer rather than an L2 specifically, the [best blockchains for cheap transfers](/best-blockchains-for-cheap-transfers/) comparison is a better starting point than this one. And if the fee itself is the puzzle, the breakdown of [how gas fees work](/crypto-gas-fees-explained/) covers why the same transaction can cost two very different amounts on two different days.

## Verdict

Base is the default answer for most people in 2026. It has the largest market share at $14.6B, the cheapest displayed fee of the major three at $0.001–0.002, and the shortest route from an exchange balance to an L2. Arbitrum One is the better venue for actual DeFi size, with $12.3B and a 6.4-day exit. OP Mainnet is for people who want the reference implementation and token-driven governance more than they want depth. Among the ZK rollups, ZKsync Era’s 3-hour exit is the standout feature, but the network holds only $227M.

None of that is settled by fees, because the top three sit within half a cent of each other. It is settled by where your money already lives, how deep the market is when you trade, and how long you can afford to wait on the way out. Get those right and the network choice follows. If you still need an on-ramp, compare [crypto exchange fees](/best-crypto-exchanges/) first.

Last updated: October 27, 2026

The post Ethereum Layer 2s Compared 2026: Base vs Arbitrum vs OP Mainnet appeared first on FogChain.