What a Layer 2 is
A Layer 2 (L2) is a separate blockchain that executes transactions off the main chain but settles its results back onto it. The main chain, the Layer 1, stays the final judge of what happened; the L2 does the heavy lifting in between. On Ethereum this means: you send a transaction on Arbitrum or Base, it costs a fraction of a cent and confirms in a second or two, and periodically the L2 writes a compressed record of everything that happened to Ethereum itself.
The reason L2s exist is simple. Ethereum mainnet processes roughly 15 to 30 transactions per second, and in busy periods a simple swap has cost 50 dollars or more. Rather than making the base layer faster (which would mean bigger blocks, more expensive nodes and fewer people able to run one), Ethereum chose to keep the base layer small and secure and push activity to layers on top. That decision, the "rollup-centric roadmap", dates from late 2020 and shapes everything below. How the base layer itself works is covered in the blockchain guide and the Ethereum guide; this one is about what sits on top.
How rollups work
Almost every serious Ethereum L2 today is a rollup. The name describes what it does: it rolls up hundreds or thousands of transactions into a batch, executes them on its own chain, and posts two things to Ethereum: the compressed transaction data, and a claim about the resulting state ("after these transactions, balances look like this"). Ethereum stores the data and, depending on the rollup type, either checks the claim mathematically or gives everyone a window to dispute it.
The important word is data. Because the full transaction data lands on Ethereum, anyone who has that data can rebuild the L2's state without asking the operator. That is the property that separates a rollup from a sidechain: a sidechain (Polygon PoS is the best-known example) has its own validators and its own security; if they fail, Ethereum cannot help you. A rollup borrows Ethereum's security for its data. A sidechain only borrows the name.
Two caveats belong right here, because "rollup" is often read as a promise it does not make. First, Ethereum keeps rollup data in blobs for only about 18 days (more on that below). Rebuilding a chain from scratch after that needs an archive of the old blobs, which rollup nodes and archive services keep, but which is outside Ethereum's own guarantee. Second, having the data and getting your money out are different things. Whether you can exit without the operator depends on what the rollup's Ethereum contracts actually allow: can you force a transaction in from L1 if the sequencer ignores you, can anyone submit a state root if the operator stops, is there an escape hatch? Arbitrum One, OP Mainnet and Base have all three. Linea, also a rollup by the data definition, has none of them as of September 2026 according to L2BEAT: no forced inclusion, whitelisted proposers only, and a takeover clause that kicks in after six months without finalised blocks. Same data model, very different exit. L2BEAT lists exactly this under "Sequencer failure" and "Proposer failure" for every chain.
Optimistic rollups: assume honesty, allow disputes
An optimistic rollup posts its state claim and assumes it is correct. Anyone who disagrees has a challenge period, about a week on Arbitrum One and OP Mainnet, to submit a fraud proof showing the claim was wrong. If nobody challenges, the state is final. If someone does and wins, the false claim is thrown out and the party that posted it loses its bond. Arbitrum One and OP Mainnet work this way, and so did Base until 2026. The cost of this design is the waiting time: withdrawing directly from Arbitrum One or OP Mainnet to Ethereum takes about seven days, because that is how long the dispute window is. Third-party bridges front you the money faster for a fee.
ZK rollups: prove correctness up front
A zero-knowledge rollup does not wait for challenges. It attaches a cryptographic validity proof to every batch, and an Ethereum contract verifies that proof before accepting the new state. If the proof checks out, the state is correct; there is nothing to dispute. Withdrawals finalise in minutes to hours instead of a week. The cost is complexity: generating those proofs takes serious computing power, the proving systems are young, and a bug in the prover or verifier is harder to spot than a bug in plain code. zkSync Era, Starknet, Linea and Scroll are ZK rollups. Polygon zkEVM was one too, until Polygon shut its sequencer down on 3 July 2026; what remains is a claims interface on Ethereum for funds that sat in ordinary wallets (EOAs) at the time, open until the end of 2027. Anything that was inside a contract on the chain, DeFi positions, multisigs, third-party bridge balances, is not covered by it. The two camps have started to merge. Since its Azul upgrade (28 May 2026) Base requires a proof for every state claim, either an attestation from a trusted hardware enclave (TEE) or a ZK proof: with one proof the claim finalises after five days, with both after one day (the five-day figure came with the Beryl upgrade in June 2026, down from seven). L2BEAT still files Base as an optimistic rollup, because a TEE-only claim can still be knocked down by a ZK challenge, but the seven-day week no longer applies to it.
| Optimistic rollup | ZK rollup | |
|---|---|---|
| How state is verified | Assumed valid; fraud proof on challenge | Validity proof checked by Ethereum |
| Withdrawal to L1 | About 7 days (Arbitrum One, OP Mainnet). Base: 5 days with one proof, 1 day with TEE and ZK together | Minutes to hours |
| Who must be honest | At least one watcher willing to challenge | The maths (plus the prover code being bug-free) |
| Examples | Arbitrum One, OP Mainnet, Base (hybrid since 2026) | zkSync Era, Starknet, Linea, Scroll |
| Value secured (L2BEAT, Sept 2026) | Base around 16 billion dollars, Arbitrum One around 12 billion | Largest ZK rollups in the low hundreds of millions |
Blobs: why L2 fees fell 90 percent in 2024
Until March 2024, rollups posted their data into ordinary Ethereum transactions and paid ordinary gas for it. Data was the biggest cost item, and L2 fees were noticeably tied to L1 congestion. The Dencun upgrade (13 March 2024) introduced blobs: a separate data lane with its own fee market, where rollup data is stored for about 18 days and then pruned, because Ethereum only needs it long enough for anyone to check or dispute the batch. After that the full history lives with the rollup's own nodes and with archive services, not on Ethereum. Blob space is cheap because it does not compete with regular transactions. L2 fees dropped by 80 to 90 percent within days.
The next step was capacity. Dencun allowed three blobs per block on average and six at most. The Fusaka upgrade (3 December 2025) shipped PeerDAS, which lets each node check only a sample of blob data instead of downloading all of it, and two follow-up forks raised the limits to a target of 14 and a maximum of 21 blobs per block by January 2026. As of September 2026, blocks carry about six blobs on average, so the ceiling sits well ahead of demand. That headroom is the reason L2 fees have stayed low even as activity grew.
The major L2s in brief
Arbitrum One
Launched in August 2021 by Offchain Labs, the first general-purpose rollup at scale and long the largest by value secured. Optimistic, with its own fraud-proof design. In February 2025 Arbitrum activated BoLD, which removed the validator allowlist: since then anyone who posts a bond can challenge an incorrect state claim, and a single honest challenger can defend the chain. Governed by the Arbitrum DAO and the ARB token. Coins in its ecosystem are listed on CoinVett's Arbitrum ecosystem page.
OP Mainnet and the OP Stack
Optimism launched publicly in December 2021 and later open-sourced its entire codebase as the OP Stack, a kit for building your own rollup. That decision turned Optimism from one chain into a family (the "Superchain") that shares code, upgrades and a security council. Its largest member, Base, left that family in 2026 (see below), which took most of the Superchain's activity with it. OP Mainnet itself is mid-sized today; the OP Stack's real footprint is the other chains built on it. Ecosystem coins: Optimism ecosystem.
Base
Coinbase's L2, launched in August 2023 on the OP Stack. Base has no token of its own and is run by a publicly listed company, which makes it the L2 most retail users touch without knowing it. It became the largest L2 by value secured in 2026. Fault proofs went live on Base mainnet on 30 October 2024, and a security council was added afterwards, taking Base to Stage 1 on L2BEAT's scale (see below). Then the ground shifted, as of September 2026: in February 2026 Base announced it was moving off the OP Stack onto its own unified software stack (the base/base repository), and in March it left Superchain governance. The Azul upgrade on 28 May 2026 put mainnet on Base's own execution client and replaced the optimistic fault-proof game with the TEE-plus-ZK multi-proof system described above; the Beryl upgrade in June cut the single-proof window to five days. Base remains an Ethereum rollup and remains Stage 1; what changed is who writes its code and how withdrawals are proven. Ecosystem coins: Base ecosystem.
zkSync Era, Starknet, Linea, Scroll
The ZK camp. Starknet uses its own virtual machine and language (Cairo), which is why it took longer to attract developers; the others aim for compatibility with existing Ethereum tooling. Technically ahead on withdrawal speed, far behind on value secured and daily users. All L2 coins tracked by CoinVett are grouped on the Layer 2 category page.
Where the trust assumptions sit
"Inherits Ethereum's security" is true in the long run and misleading in the short run. Every major L2 still has a few places where you trust people rather than code.
- The sequencer. The component that orders transactions and produces L2 blocks. On every major rollup it is a single operator (Offchain Labs, Optimism, Coinbase). If it goes down, the chain stops producing blocks. Base halted for 33 minutes on 5 August 2025 when a failover to a backup sequencer did not work. Whether your funds are stuck in such an outage depends on the chain. Arbitrum One, OP Mainnet and Base let you submit a transaction to a contract on Ethereum that the rollup must include within a set delay (up to a day on Arbitrum, up to twelve hours on Base), and from there the normal withdrawal path with its proof window applies; slow, and few users know it exists, but it works without the operator. Not every rollup offers it: Linea has no such mechanism as of September 2026, so if its sequencer stops or censors you, your exit waits until the operator returns or is replaced.
- Upgrade keys. The L2's contracts on Ethereum can be upgraded. Who holds the keys, how many signatures are needed and whether there is a delay decides how much a rollup can change under your feet.
- The proof system. Optimistic rollups need at least one honest party watching and willing to challenge. ZK rollups need the prover and verifier code to be correct.
L2BEAT, an independent research site, grades rollups on exactly these points using a Stages framework proposed by Vitalik Buterin in 2022. Stage 0 means the operators are effectively in control (training wheels on). Stage 1 means a working proof system is live, users can exit without the operator's help, and only a security council with a 75 percent threshold can override. Stage 2 means the code is in charge, with at least 30 days' notice before any upgrade. Arbitrum One, OP Mainnet and Base reached Stage 1 between 2024 and 2025. As of 2026, no large general-purpose rollup has reached Stage 2. Check the current grade on L2BEAT before moving significant funds; it changes.
Bridging to an L2 and back
To use an L2 you move assets there through a bridge. The canonical bridge is the rollup's own contract on Ethereum: you deposit ETH or tokens, they are locked on L1 and credited on L2. It is the safest route because it carries the rollup's own security, but withdrawals back to L1 wait for the rollup's proof window: about a week on Arbitrum One and OP Mainnet, one to five days on Base. Third-party bridges (Across, Hop, Stargate and others) offer near-instant transfers by fronting liquidity, at a fee and with their own smart contract risk. Bridges have been the single largest source of losses in crypto: the Ronin bridge (625 million dollars, March 2022) and Wormhole (325 million, February 2022) were not L2 canonical bridges, but they show what a bridge bug costs. Contract-level risks of the tokens you bridge can be checked with the Token Safety Check; the bridge contract itself is a separate question.
The Lightning Network: Bitcoin's different answer
Bitcoin has an L2 too, but it works nothing like a rollup. The Lightning Network is a network of payment channels. Two parties lock bitcoin in a shared on-chain transaction, then exchange signed balance updates off-chain as often as they like, for free and instantly. Only the opening and closing of the channel touch the Bitcoin blockchain. Payments can be routed across several channels, so you do not need a direct channel with everyone you pay.
Lightning does not post data to Bitcoin and is not a general-purpose smart-contract platform; it is built purely for payments. It does, however, rest on Bitcoin's own contract language: channels are secured with Bitcoin Script, using timelocks and hash-locked contracts (HTLCs) so that either party can settle on-chain if the other cheats or disappears. Capacity is limited by what is locked in channels: about 4,900 BTC in public channels across roughly 17,000 nodes and 41,000 channels as of mid-2026, plus an unknown amount in private channels. Both parties have to be online to receive, and a channel's balance caps how much can flow through it. In practice most users today access Lightning through a wallet or exchange that manages channels for them, which reintroduces a custodian. It is a payment rail, not a general-purpose platform, and comparing its numbers to Arbitrum's misses the point of both.
Frequently asked questions
Is my ETH on an L2 "real" ETH?
It is a claim on ETH locked in the rollup's bridge contract on Ethereum, redeemable through the canonical bridge. For a rollup with a working proof system, that claim is backed by Ethereum's own rules. For a sidechain, it is backed by the sidechain's validators. Know which one you are on.
Why does withdrawing from an optimistic rollup take days?
Because the rollup gives everyone time to challenge a state claim before it becomes final, and the wait is the security mechanism, not a bug. How long depends on the chain, as of September 2026: about seven days on Arbitrum One and OP Mainnet. On Base every claim now carries a proof, so it is five days with a single TEE or ZK proof and one day when both agree. Third-party bridges are faster because someone else takes on the waiting for a fee.
Do L2s have their own tokens, and do I need them?
Some do (ARB, OP, ZK, STRK), used for governance. Whether you need them to use the chain depends on the chain. On Arbitrum One, OP Mainnet, Base, zkSync Era, Linea and Scroll, gas is paid in ETH. Starknet is the exception: since protocol version 0.14 its fee limits are set in STRK and the protocol charges fees in STRK, so a wallet paying its own way needs some. Starknet also has paymasters, contracts that pay the STRK on your behalf: a dApp can sponsor the fee outright, or you pay in another supported token and the paymaster converts. Whether that path is open to you depends on the wallet and the app, not on the protocol. Base has no token at all. Holding an L2's governance token is a separate decision from using the L2, and this guide does not address it.
Which Layer 2 is the best?
Depends on what you need. Base and Arbitrum have by far the most users, liquidity and applications. ZK rollups have faster exits. L2BEAT's Stage grade tells you how much you still rely on the operator. None of that is a statement about any token's price, and CoinVett does not rank chains as investments.
Notice: This page describes how Layer 2 networks work and where their risks sit. It is not a recommendation to use any particular chain or to hold any token.