Optimistic vs zk-rollups: how Ethereum's layer 2s compare

An optimistic rollup and a zero-knowledge rollup both run transactions off Ethereum, bundle them together, and post the results back to the main chain. The difference is trust. An optimistic rollup assumes each batch is valid and lets anyone challenge it inside a set window. A zk-rollup proves each batch is valid before Ethereum will accept it at all.

Both are layer 2 networks, usually shortened to L2. Ethereum now treats them as its main way to scale. The idea is plain. Do the heavy work somewhere cheaper, then settle the result on the chain everyone already trusts. Which of the two designs a network uses decides how fast you get your money out, how much a transaction costs, and what can go wrong along the way.

What a rollup actually does

A rollup takes hundreds of separate transactions and treats them as one. An operator, usually a single computer called a sequencer, gathers your trades and transfers off-chain. It puts them in order, runs them, and writes a compressed record to Ethereum. That record is the part Ethereum keeps. Spreading one fixed cost across hundreds of users is what makes each transaction cheap.

Ethereum itself does not re-run those transactions. It stores the data and a set of smart contracts that track the rollup's balances and state. This is the bargain every rollup strikes. You get Ethereum's security for your data without paying Ethereum's full price on every click. Lower fees are the whole reason rollups exist, and they are why these networks now carry most of Ethereum's day-to-day activity.

Picture a shared taxi instead of twenty separate cabs. Everyone rides together, the fare splits twenty ways, and only the finished trip gets logged. A rollup does the same with transactions. The sequencer is the driver, Ethereum is the logbook, and your fee is a share of one entry rather than the whole ride.

That data part matters more than it sounds. Ethereum added a cheaper storage slot, called a blob, in an upgrade known as EIP-4844. Rollups post their batches into blobs now. Ethereum drops those blobs after about 18 days, while the older method, calldata, stayed on-chain forever. That one change cut rollup fees hard and is why a layer 2 transaction can cost a few cents.

Both kinds of rollup still lean on Ethereum for the hard part. The rollup does the computing, but Ethereum holds the data and settles who owns what. That is the line between a rollup and a plain sidechain. A sidechain runs on its own and asks you to trust its own validators. A rollup posts enough to Ethereum that anyone could, in theory, rebuild its accounts from scratch. Lose that data link and the security argument falls apart. Ethereum's own documentation says rollups inherit the main chain's security for exactly this reason.

Optimistic rollups: trust first, check later

An optimistic rollup is optimistic by name and by design. It assumes the operator posted honest data and publishes no proof that the math is correct. Ethereum takes the batch at face value. Then it waits.

Its safety net is a fraud proof. For a fixed period, anyone watching can flag a bad batch and force the network to re-check the disputed transactions directly on Ethereum. Operators post a bonded deposit first. If a challenge succeeds, the protocol reverses the bad block and the operator loses that deposit. Arbitrum, Optimism, and Base all work this way, and together they hold most layer 2 activity.

This waiting is where the famous week-long delay comes from. Funds leaving the rollup for Ethereum have to sit through the challenge window before they land. Arbitrum's documentation sets that window at 45,818 Ethereum blocks, which it puts at about 6.4 days. Ethereum's own docs round it to roughly seven days. Either way, a direct withdrawal takes about a week.

Zero-knowledge rollups: prove it up front

A zk-rollup flips the order around. Instead of waiting for someone to catch a problem, it submits a cryptographic validity proof with every batch. Ethereum checks the proof. Only if the math holds does the update go through. Ethereum's documentation calls this proof "a cryptographic assurance that the state-change proposed by the rollup is really the result."

Two kinds of proof do this job. A ZK-SNARK is small and fast to verify but needs a one-time trusted setup to get started. A ZK-STARK needs no trusted setup and holds up against future quantum computers, at the cost of a bigger proof. Checking one on Ethereum still burns real gas, around 500,000 units by Ethereum's own estimate. Networks such as zkSync, Starknet, Scroll, Linea, and Polygon zkEVM take this path.

At the exit, the payoff is clear. Because Ethereum has already verified the proof, there is no challenge window to sit through. Ethereum's docs say it flatly: "there are no delays when moving funds from a ZK-rollup to Ethereum." Your withdrawal clears in the time it takes the proof to post and verify. Hours, not a week.

Optimistic vs zk: the trade-offs that matter

Start with how fast you can withdraw, because that is what an ordinary user feels first. Optimistic rollups mean about a week for a direct exit. Zk-rollups mean hours. If you shift large sums between a layer 2 and Ethereum often, that gap is the single biggest practical difference between them.

Next comes developer effort. Optimistic rollups were easier to build on for years, because they copy Ethereum's execution closely and existing apps move over with few changes. Zk-rollups were harder, since proving general Ethereum computation is tough. That gap has shrunk. A zkEVM rebuilds Ethereum's instructions inside the proof system, so ordinary contracts can run in a provable form.

Then there is cost and complexity under the hood. Producing validity proofs is heavy work and needs specialized hardware. Optimistic rollups avoid that expense but carry the week-long exit and the need for honest watchers to be present. One point cuts against the easy conclusion. Faster proofs do not make a network safer by themselves. A bug buried in a proving circuit could wave an invalid state straight through, and no challenge window exists to catch it afterward.

Neither design wins in the abstract. A zk-rollup suits anyone who values quick exits and strong up-front guarantees. An optimistic rollup suits builders who want Ethereum-style tooling with the least friction. No single model has been crowned. The newest networks increasingly choose the zk path, while the largest optimistic rollups still hold the most value today.

How your money leaves a layer 2

Direct withdrawals follow the rules above. On an optimistic rollup you wait out the challenge period. On a zk-rollup you wait for the proof to verify. Most people never wait that long, because a faster route exists.

Liquidity providers and bridges will hand you the money on Ethereum right away, for a fee, then claim the slow withdrawal themselves later. It works like a check-cashing shop. Pay a small cut, skip the hold. The catch is that the risk moves to the bridge, and bridges have been drained for large sums before, so a fast exit is not a free one.

There is a quieter risk under both designs. Most rollups still run a single sequencer. If it stalls, your transactions stop until it restarts or a backup method, sometimes called an escape hatch, lets you exit straight to Ethereum. Plans to decentralize sequencers exist. On the biggest networks they have not fully shipped. L2BEAT, an independent tracker, still ranks the largest rollups at Stage 0 or Stage 1, short of the full Stage 2 that would hand the most control back to users.

What to watch

Watch the exit route you actually use, not the label on the network. A fast bridge turns a week-long optimistic withdrawal into minutes, but it hands your funds to a third party. A direct withdrawal is slower and safer. Pick the one that matches the amount at stake.

Where the data goes matters just as much. A true rollup posts its transaction data to Ethereum, which is what lets anyone rebuild the chain if the operator disappears. Some networks keep that data off-chain to cut costs and are not really rollups at all. Ethereum's docs also warn that publishing data on-chain "is not a privacy mechanism," so never treat a rollup as private.

The last thing to watch is which networks survive. On 4 November 2026, Lisk shut its own Ethereum layer 2 after years of running it, a reminder that an L2 is a business as much as a piece of software. Neither camp, optimistic or zk, is going away. Which networks win inside each one is still being settled.