Gas on Ethereum still stings, and that’s the only reason any of these side chains exist. Ethereum’s own figures put the typical fee on an Ethereum-backed network at $0.0020 at the time of writing (September 2026), while the main chain sits at $0.034 for the same job.
Here’s the quick version. Go to Arbitrum One when you’re doing DeFi and you’d like a chain that anybody is allowed to police, or to Base when your coins already live at Coinbase. Pick one of the proof-based chains, meaning Linea, Scroll, ZKsync Era or Starknet, if the thought of waiting a week to get your money home makes you twitch.
Below we go through the idea itself, then each network, then the bits that decide it for most people: who holds the keys, what you’re charged, and how long you’re kept waiting on the way out. We’ve also given Polygon zkEVM a section, because it shut down in July 2026 and half the internet hasn’t noticed. Nothing here is invented; every figure comes from the chain’s own docs, and we’ve linked them.
Table of contents
Every layer 2 on our shortlist, in one table
Ethereum keeps its own roster of layer 2 networks, sorted by market share, with what an average transaction costs on each one. Those totals shift by the day, so what follows is a photograph taken in September 2026 and not a league table carved in stone.
Best Ethereum Layer 2 Networks: Comparison Chart | |
|---|---|
| <strong>How it proves itself</strong> | Arbitrum One calls itself a trustless rollup <br> Base and OP Mainnet are optimistic rollups with fault proofs <br> ZKsync Era, Starknet, Linea and Scroll prove every batch |
| <strong>Whose hands it’s in</strong> | Arbitrum One answers to the Arbitrum DAO <br> Base needs Coinbase plus an 11-seat Security Council <br> OP Mainnet belongs to the Optimism Collective <br> The ZK chains are still run by the teams that built them |
| <strong>The number you type into a wallet</strong> | Arbitrum One: 42161 <br> Base: 8453 <br> OP Mainnet: 10 <br> ZKsync Era: 324 <br> Linea: 59144 <br> Scroll: 534352 |
| <strong>What pays for gas</strong> | ETH on Arbitrum One, Base, OP Mainnet, ZKsync Era, Linea and Scroll <br> Starknet wants STRK instead |
| <strong>Typical cost of a transaction</strong> | Base: $0.002 <br> Arbitrum One: $0.005 <br> Optimism: $0.00 <br> Starknet: $0.005 <br> Scroll: $0.002 <br> Linea: $0.024 |
| <strong>Size, by market share</strong> | Base: $14.6B <br> Arbitrum One: $12.3B <br> Optimism: $1.65B <br> Starknet: $389M <br> Linea: $352M <br> ZKsync Era: $227M <br> Scroll: $45.7M |
| <strong>Getting money back to Ethereum</strong> | Base and OP Mainnet make you wait 7 days <br> Arbitrum One runs a dispute window of 45818 blocks, roughly 6.4 days <br> ZKsync Era holds you for a 3 hour minimum |
| <strong>How often blocks land</strong> | Base and OP Mainnet: every 2 seconds <br> Arbitrum One has no fixed block time at all <br> Starknet closes a block within 9.5 seconds |
| <strong>How cheap fees can get</strong> | Base stops at 0.005 gwei <br> Arbitrum One won’t go under 0.02 gwei <br> Linea settles around 7 wei |
| <strong>Where you watch your transaction</strong> | Arbitrum One: arbiscan.io <br> Base: basescan.org <br> OP Mainnet: explorer.optimism.io <br> ZKsync Era: explorer.zksync.io <br> Linea: lineascan.build <br> Scroll: scrollscan.com |
| <strong>It suits you if</strong> | Arbitrum One: you’re deep in DeFi <br> Base: you bank at Coinbase and push stablecoins around <br> OP Mainnet: you want a vote <br> Linea and Scroll: you hate slow exits |
So what is an Ethereum layer 2, really?
Your transaction runs on a separate chain, and a copy of the data goes back to Ethereum afterwards. That second step is what buys you Ethereum’s guarantees at a fraction of Ethereum’s price, and it’s where the fee gap above comes from.
Where the data ends up is the question worth asking. Ethereum’s own pages put it as “whether the network stores its data on the Ethereum main network”, and the chains that do earn the name rollup. Anything storing its data elsewhere isn’t really an extension of Ethereum, and it’s leaning on its own security instead.
How a rollup actually guards your money
- An optimistic rollup takes your transaction at face value and leaves a window open for anyone to object. On Base and OP Mainnet that window lasts 7 days, which is the whole reason a withdrawal drags on for a week.
- A validity rollup does the maths first and hands Ethereum a proof. Starknet describes itself as exactly that, using STARK-based zero-knowledge proofs to pack thousands of transactions together off-chain.
- Gas just means the toll you hand over for using a chain. Nearly everything here takes it in ETH, so you don’t have to go shopping for a second coin before you start, and only Starknet breaks the pattern by asking for STRK.
- All of these chains post their data to Ethereum, which means that if the operator vanished tomorrow, the state could be rebuilt from Ethereum on its own.
Arbitrum One is where DeFi actually lives
The Arbitrum docs bill the project as the finance-native platform providing infrastructure for applications, tokenization and dedicated chains. Arbitrum One is the one you’ll end up using. It answers to chain ID 42161, takes its gas in ETH, and turns up on Arbiscan and Blockscout alike.
Documentation for the project describes Arbitrum One as implementing the purely trustless rollup protocol. Over on chain ID 42170, Arbitrum Nova runs the mostly trustless AnyTrust protocol, which leans on a Data Availability Committee. Cheaper to use, one more group you’re asked to believe in.
We’ve set Arbitrum against its rivals twice already, in Arbitrum vs Optimism and Arbitrum vs Polygon. Both times the technical distance turned out narrower than the marketing wanted us to believe.
What Arbitrum One gets right
- BoLD is short for Bounded Liquidity Delay, it’s switched on across Arbitrum One, Arbitrum Nova and Arbitrum Sepolia, and it throws validation open to anybody instead of a vetted list.
- With Stylus you can write EVM-compatible contracts in Rust, C and anything else that compiles to Wasm. Handy when nobody on your team has ever written Solidity.
- Should the sequencer decide to ignore you, the force-include period of 5760 blocks lets you push the transaction through yourself. That’s 24 hours of waiting.
- Blocks cap out at 32,000,000 gas, and the price of gas won’t fall below 0.02 gwei even on a dead quiet afternoon.
- Ethereum’s roster has Arbitrum One in second place on market share, at $12.3B in September 2026, and that size is why practically every DeFi app supports it.
Base sits one click away from Coinbase
Base came out of Coinbase, and its documentation pitches it as the blockchain for global finance. The docs promise transactions that settle in under a second for less than a cent. Ethereum’s figures support the cost half of that promise, at least.
You’ll want chain ID 8453 and the RPC at mainnet.base.org, with ETH as the currency and basescan.org for looking things up. Developers testing first should use Base Sepolia, which is chain ID 84532.
On Ethereum’s own list Base takes the top spot for market share at $14.6B, and an average transaction there ran $0.002 in September 2026. If your coins sit at Coinbase already, it’s also the shortest trip onchain you’ll find.
What Base gets right
- A Flashblock gets built every 200ms, which drags the effective block time down from 2 seconds to 200 milliseconds by preconfirming. In a wallet, a swap just feels done.
- The team claims the chain has held multiple bursts above 5,000 TPS, which a full block gas budget near 400M makes plausible.
- Its base fee bottoms out at 5,000,000 wei, otherwise written as 0.005 gwei. Base does the arithmetic for you too, landing on roughly $0.002 for a normal 200,000 gas transaction with ETH at $2000.
- Fault proofs went permissionless back in October 2024, and by April 2025 the chain had reached what’s called Stage 1 decentralization.
- Proving and finalizing a withdrawal is open to anyone, not just whoever kicked it off, so an exit nobody relays isn’t lost forever.
OP Mainnet hands the keys to two voting houses
This is Optimism’s original chain. It runs at chain ID 10, with mainnet.optimism.io for the RPC and explorer.optimism.io for looking up transactions, and a fresh block arrives every two seconds whether there’s anything in it or not.
Optimism calls the OP Stack the standardized, shared, open-source development stack behind Optimism, built so anyone can spin up a production-ready layer 2 of their own. Base is built on it as well, which is why using the two chains feels nearly identical.
Voting is what really sets it apart. Decisions go through two houses of the Optimism Collective: a Token House made up of OP holders and their delegates, and a Citizens’ House working on one member, one vote.
Where OP Mainnet earns its place
- June 10, 2024 is the date fault proofs joined the OP Stack and went live here, and challenges stopped needing anyone’s permission from then on.
- Protocol upgrades come from the Developer Advisory Board, and either house can veto them. The Citizens’ House also picks who sits on that board.
- Running a transaction costs you precisely what the same transaction would cost in gas on Ethereum, and then an L1 data fee lands on top.
- Batches travel to Ethereum as EIP-4844 blobs, compressed on the way, and that compression is what keeps the Ethereum share of your bill small.
ZKsync Era, Starknet, Linea and Scroll: the proof crowd
This group settles with maths rather than with a waiting period. You won’t notice the difference while you’re using them; you’ll notice it on the way out. Our roundup of the top ZK roll-up projects covers the wider field.
Built on the ZK Stack and running the EraVM, ZKsync Era sits at chain ID 324 and uses ETH as its base token. Because account abstraction is baked in, paymasters can let you settle fees in an ERC-20 such as USDC, which means you can transact without owning any ETH at all.
Leaving a ZKsync chain means clearing a 3 hour minimum delay, set by ZIP-4. To its credit the team says plainly that the delay isn’t a promise of finalization within 3 hours, since how long it really takes depends on traffic and on proof generation.
Starknet is the odd one out, and deliberately so. The docs call it a decentralized, permissionless validity rollup resting on STARK-based zero-knowledge proofs, contracts are written in Cairo instead of Solidity, and the fee comes out of your STRK rather than your ETH. A block closes inside 9.5 seconds, or after 2 seconds once the mempool empties.
Over on Linea, a zkEVM chain at chain ID 59144, gas is ETH again and Lineascan is the explorer. The docs advertise soft finality within 1s and a base fee that settles at 7 wei. Ethereum’s roster, awkwardly, gives Linea the priciest average of the four at $0.024.
Scroll is another zkEVM, this one at chain ID 534352, again with ETH for gas and scrollscan.com for lookups. OpenVM does the proving, and your transaction counts as final once the rollup node puts a validity proof on Ethereum. At $45.7M of market share it’s the smallest name here, so expect fewer apps and thinner liquidity than the rest.
Polygon zkEVM got switched off
You’ll still find Polygon zkEVM on best-layer-2 roundups that nobody has bothered to revisit. The chain is dead. Polygon’s own page says that as of July 3, 2026, the Polygon zkEVM Mainnet Beta sequencer has been sunset.
No blocks are being produced any more, and the Agglayer Bridge won’t process a withdrawal from it. Polygon published the timeline back in June 2025, so there were 12 months of warning before anything stopped.
Anyone who kept coins there in a self-custodied wallet can still pull them out on Ethereum through the Polygon zkEVM Claims interface, which stays open until December 31, 2027. Money sitting inside a smart contract is the painful case: DeFi positions, multisigs and bridge contracts can’t be rescued that route. Polygon’s PoS chain carries on quite separately, and we walk through it in our guide to bridging from Ethereum to Polygon.
The part each layer 2 asks you to take on faith
Not one of these chains has given up its single sequencer yet. Optimism writes it down without dressing it up, saying the Optimism Foundation runs the only block producer on OP Mainnet, and everybody else on this list is in the same spot.
What separates them is the emergency exit. Arbitrum will let you shove a transaction through after 5760 blocks, which is a day. Base and OP Mainnet let any passer-by prove and finalize your withdrawal once the challenge period has run out.
Upgrade keys deserve as much of your attention as the proof system does, and honestly they get far less of it. Base requires a 2-of-2 multisig, with Coinbase’s signers at 3-of-6 on one side and a Security Council of 11 independent entities at 8-of-11 on the other. Arbitrum One passes upgrades to its DAO, while OP Mainnet lets either voting house block them.
Proof-based chains swap one delay for a different one. ZKsync parks your withdrawal behind a time lock for at least 3 hours, buying the team a window to look into anything odd, such as a sudden rush of money heading for the door.
None of that matters next to the mistake people actually make. Choose the wrong network, or paste the wrong address, and the money is gone with nobody able to claw it back. Look at that network selector twice, and if the string you’re pasting means nothing to you, our explainer on what a wallet address is is worth five minutes.
Got a balance you’d hate to lose? Keep the keys on a device rather than a browser extension, and our rundown of the best hardware wallets for MetaMask works for every EVM chain in this article.
What you’ll pay, and how long you’ll wait
Two things make up your bill: the cost of running the transaction on the layer 2, and the cost of writing it to Ethereum. Base spells that split out in its docs, and the Ethereum half is usually the bigger one.
Don’t read a floor as a forecast. Base won’t drop under 0.005 gwei, Arbitrum One holds its line at 0.02 gwei, Linea settles near 7 wei, and yet the amount leaving your wallet still follows whatever blob space costs on Ethereum that day.
| Network | Typical transaction | How quickly blocks land | Getting out to Ethereum |
|---|---|---|---|
| Base | $0.002 | Every 2 seconds, preconfirmed at 200ms | A week |
| Arbitrum One | $0.005 | Whenever there are transactions to sequence | Roughly 6.4 days |
| OP Mainnet | $0.00 | Every 2 seconds | A week |
| Starknet | $0.005 | Closed within 9.5 seconds | Settled by proof |
| Linea | $0.024 | Soft finality inside 1s | Settled by proof |
| Scroll | $0.002 | The docs don’t say | Settled by proof |
Yes, Linea’s $0.024 average is about ten times what the cheapest chains here charge. Set it beside the $0.034 that Ethereum itself averages, though, and the difference stops mattering, which is why we wouldn’t pick a chain on fees alone.
The good and the bad on the top two
Arbitrum One: what we like, what we don’t
Base: the upside and the catch
So which layer 2 is right for you?
Coinbase customer who mainly shuffles stablecoins around? Base. It’s the biggest of them by market share, its $0.002 average is as cheap as this list gets, and preconfirmations show up in about 200ms.
Anyone spending real time in DeFi, and bothered by the idea of a closed validator set, should be on Arbitrum One instead. BoLD runs there, upgrades belong to the DAO, and at $12.3B of market share the apps you want are already deployed.
Can’t stand the week-long exit? Then go proof-based. Linea and Scroll close out with proofs, and ZKsync Era turns that week into a 3 hour minimum.
Layer 2 questions we get asked most
Which Ethereum layer 2 network is the biggest?
Base, by market share. Ethereum’s roster had it in front at $14.6B in September 2026, with Arbitrum One next at $12.3B and Optimism third on $1.65B.
Where will my transactions cost the least?
Base and Scroll tie at the bottom of Ethereum’s list, both averaging $0.002. Arbitrum One and Starknet come in at $0.005, and Linea is dearest at $0.024. Every one of those is loose change beside the $0.034 an average Ethereum transaction costs.
Why do I have to wait a week to withdraw?
Because an optimistic rollup treats your transaction as valid unless somebody objects, and it has to leave time for that objection. Base and OP Mainnet make the window 7 days. Arbitrum One uses a dispute window of 45818 blocks, or about 6.4 days.
Will I need a separate token for every layer 2?
Rarely. ETH pays the gas on Arbitrum One, Base, OP Mainnet, ZKsync Era, Linea and Scroll. Starknet is the one that asks for something else, charging its fees in STRK.
Can I still use Polygon zkEVM?
No. Its Mainnet Beta sequencer was sunset on July 3, 2026, and nothing is producing blocks there now. If you left coins in a self-custodied wallet, you’ve got until December 31, 2027 to claim them on Ethereum.
Are these layer 2 networks genuinely decentralized?
Halfway there. Each chain on this list is still sequenced by one party, and Optimism says outright that the Optimism Foundation runs the only block producer on OP Mainnet. Challenging a state or proving a batch is open to everyone; producing the blocks is not, and that’s the gap we’d like closed.
Does my hardware wallet work on a layer 2?
Yes, on the EVM ones. Add the chain ID and RPC and any wallet that already handles Ethereum handles these too. Starknet is the exception again, since Cairo contracts need wallets built for Starknet.
Our verdict on the best Ethereum layer 2 networks
Most people will end up on Base or Arbitrum One, and what separates the two is governance rather than clever engineering. Base wins on size, on average cost, and on how easily you reach it from an exchange. Arbitrum One reports to a DAO and lets outsiders validate through BoLD, which we’d weigh heavily if we were keeping a serious balance onchain.
Behind them sit the proof-based chains, and they’re the ones we find more interesting. Linea, Scroll, ZKsync Era and Starknet close their books with cryptography instead of a week-long objection window, though Starknet demands the most from you, with fees in STRK and contracts written in Cairo rather than Solidity.
This article is for general information only and is not financial, legal, or investment advice. Prices and features change; check the vendor’s official page before buying.
