Optimistic vs. ZK rollups: what's the difference?

Rollups are Ethereum's scaling solution: they execute transactions off-chain and post compressed data plus a proof of correctness back to Ethereum. The two families — optimistic and zero-knowledge — differ in what that proof looks like. Optimistic rollups assume everything is fine and let anyone challenge a bad batch during a window. ZK rollups prove every batch is correct using a zero-knowledge proof posted on-chain. Both models work. They have different properties, and the difference shapes every downstream tradeoff.

How each works

An optimistic rollup (Arbitrum, Optimism, Base, Blast) posts the compressed transaction data to Ethereum and says, in effect, "here is the new state root, assume this is right." A seven-day challenge window opens. During that window, any observer who thinks the batch was invalid can submit a fraud proof: a specific step of execution they claim is wrong. Ethereum adjudicates the dispute by re-executing the contested step. If the challenger is right, the batch is reverted and the submitter loses their bond. If the challenger is wrong, they lose their bond.

A ZK rollup (zkSync, StarkNet, Polygon zkEVM, Scroll, Linea) does not assume anything. Every batch comes with a zero-knowledge proof (see the ZK-proof explainer) that the batch was executed according to the rules. Ethereum verifies the proof in milliseconds. If the proof is valid, the new state is accepted immediately. If it is invalid, the batch is rejected.

Finality and withdrawals

The biggest practical difference is withdrawal time. With an optimistic rollup, funds moving from the rollup back to Ethereum have to wait out the challenge window — seven days on Arbitrum and Optimism — because only after that window does the mainnet know for certain the batch was valid.

Users rarely wait seven days. Third-party bridges (Across, Hop, Stargate, Orbiter) will front them their funds on mainnet against their pending rollup withdrawal, taking the withdrawal-timing risk themselves in exchange for a small fee. The user experience is "fast bridging" in exchange for paying an intermediary.

ZK rollups have no such delay. Once a proof lands on Ethereum, the withdrawal is final. In practice, proof generation takes some time (minutes to hours depending on the system), but that delay is much shorter than the optimistic challenge window, and it does not require any third-party trust to work around.

EVM compatibility

Optimistic rollups have had an easier time supporting the full Ethereum Virtual Machine. Because the fraud proof is resolved by Ethereum replaying a contested step, the rollup can execute nearly anything the EVM can. Arbitrum and Optimism have been broadly EVM-equivalent for years.

ZK rollups have struggled with the EVM. The EVM is not zk-friendly — certain opcodes and data structures are expensive to prove. Early ZK rollups (zkSync Lite, StarkEx applications) were not EVM-compatible at all; they required custom languages like Cairo. Modern zkEVM rollups (zkSync Era, StarkNet v11 with Kakarot, Polygon zkEVM, Scroll, Linea) approximate the EVM at different levels of faithfulness. Scroll and Linea are closer to full bytecode-level equivalence; Polygon zkEVM is similar; zkSync Era has some compatibility differences that require care.

The gap is closing. The tradeoff is shrinking every year as proof systems improve.

Data availability and the Dencun upgrade

Both rollup types post data to Ethereum for anyone to reconstruct the rollup's state. Before March 2024, this data was posted as calldata, which was expensive. The Dencun upgrade introduced blobs (EIP-4844) — a cheaper data-availability lane designed specifically for rollups. Blob space is pruned after about 18 days, which is fine because rollups only need the data long enough for verification.

Dencun cut rollup fees by roughly a factor of ten to twenty. A swap on Arbitrum or Base now costs a few cents. The next upgrade (Pectra and beyond) will continue scaling blob capacity.

Security models

A rollup's security depends on several things. First, the proof system itself: a bug in the verifier contract, or in a zk circuit, can accept bad batches. Second, the sequencer: every active rollup has a centralized sequencer that orders transactions. A malicious or failing sequencer can censor or reorder, though not steal funds (users can always force-include transactions through Ethereum). Third, governance: most rollups have a "security council" multisig that can upgrade the rollup contracts, which is a trust assumption in the team.

The L2Beat project tracks these dimensions carefully and rates each rollup by how many guardrails are still the team's discretion vs. the protocol's. Most rollups are still "stage 0" (training wheels), meaning the team can upgrade fast; fewer are "stage 1" or "stage 2" (progressively more decentralized).

Which to choose

For users: the answer is usually whichever rollup the application they want to use is on. Arbitrum and Base have the deepest DeFi ecosystems. Optimism has the Superchain vision and OP Stack. StarkNet and zkSync have the largest ZK-native ecosystems. Performance differences between the two families are mostly invisible for retail-scale usage.

For builders: the choice is more nuanced. Optimistic rollups are mature and EVM-equivalent; ZK rollups are newer, occasionally EVM-different, but finality is better. Apps that need fast exits (cross-chain arbitrage, high-frequency trading) are more natural on ZK chains. Apps that need full Solidity compatibility and mature tooling are still often better on optimistic.

Why it matters

Rollups are how Ethereum scales. The decision between optimistic and ZK is not a winner-take-all — both types will likely coexist indefinitely, optimizing for different workloads — but the long-term trajectory favors ZK as proof systems improve. Within a few years, the cost advantage and UX advantage of instant finality will likely make ZK the default for new rollups, while existing optimistic rollups (which have the largest ecosystems) coexist and gradually integrate ZK techniques. The distinction will keep mattering, and the details will keep shifting, for at least the next decade.