Ethereum L2s in 2026: Arbitrum, Optimism, Base and zkSync Compared

Why L2s exist

ethereum.org puts Ethereum mainnet's capacity at roughly 15 transactions per second, and fees rise whenever demand for block space outstrips that. This is too slow and expensive for most consumer applications.

Layer 2 rollups solve this by executing transactions off the main chain, bundling hundreds of them together and posting the result back to Ethereum. Because many transactions share the cost of one mainnet transaction, ethereum.org says fees are "massively reduced", while Ethereum remains the settlement layer.

A second change made L2s cheaper again. Proto-danksharding (EIP-4844), which went live in the Dencun upgrade in March 2024, lets rollups post their data as "blobs" that Ethereum nodes delete after about 18 days, instead of as permanent calldata that every node keeps forever (ethereum.org).

Worked example (illustrative figures). Suppose a rollup pays the equivalent of 5 dollars to post one batch to Ethereum, and that batch carries 1,000 user transactions. Each user's share of the Ethereum cost is 5 / 1,000 = 0.005 dollars, plus the L2's own execution fee. The same 1,000 transfers made individually on mainnet would each pay the full mainnet fee. That is the whole economic case for rollups: the more transactions per batch, the smaller each share. Live fees change with demand, so check a fee tracker or your wallet's estimate rather than any figure in an article.

Optimistic vs ZK rollups

• Optimistic rollups (Arbitrum, OP Mainnet, Base) assume batches are valid but let anyone challenge an invalid one during a challenge window. Arbitrum's documentation gives the challenge period as 6.4 days (Arbitrum docs); OP Stack withdrawals wait for "the fault challenge period (7 days on mainnet)" (Optimism docs). That window is why a native withdrawal back to Ethereum takes about a week.
• ZK (validity) rollups (ZKsync Era) post a cryptographic proof that the batch is correct, so they do not need a challenge window for correctness.

The major L2s in 2026

L2BEAT rates rollups by "stage", a measure of how far they have moved from operator control to code control: Stage 0 is "fully controlled by few entities", Stage 2 is "fully controlled by code", and Stage 1 sits between, with a Security Council that can still intervene (L2BEAT stages). L2BEAT stresses that stages measure decentralisation, not overall security.

Source for type and stage: L2BEAT. Stages change as projects upgrade, so check the live page before relying on one.

Arbitrum (ARB) One of the most widely used general-purpose L2s. Optimistic rollup architecture. Home to established DeFi apps such as GMX and Camelot. Fees vary with demand and Ethereum's data costs, so check a live fee tracker or your wallet's estimate.

Optimism (OP) Pioneer of the optimistic rollup model. Coinbase built Base on its OP Stack, and other chains use the same code.

Base (Coinbase) Built on the OP Stack, operated by Coinbase. Base launched without a network token; in September 2025 it said it was beginning to explore one, while Coinbase said it had no definitive plans; check Base's own channels before trusting anything sold as a "Base token". Consumer apps such as Friend.tech launched here.

ZKsync Era ZK rollup architecture: mathematically provable transaction validity rather than fraud proofs. L2BEAT currently places it at Stage 0, so the proof system is only part of the trust picture; who can upgrade the contracts matters too.

Polygon zkEVM (closed) Polygon Labs sunset the Polygon zkEVM Mainnet Beta sequencer on 1 July 2026, after telling users to bridge their funds back to Ethereum. It is no longer an option for new activity; Polygon's other networks are separate chains with their own trade-offs.

Which L2 for what

• DeFi and yield: Arbitrum (large, established DeFi ecosystem); our DeFi beginner's guide explains why small positions belong on an L2
• Consumer apps and social: Base (Coinbase rails, easy onboarding)
• Gaming: whichever chain the game you want to play actually runs on
• ZK rollup: ZKsync Era

Bridging: the withdrawal arithmetic and the risk

Moving assets between Ethereum and an L2 uses a bridge. The native bridge is the rollup's own contract: deposits arrive in minutes, but withdrawals from an optimistic rollup wait out the challenge period. Third-party fast bridges such as Across and Stargate front you the funds on the destination chain for a fee and take the slow exit themselves.

Worked example (illustrative fee). You want 2,000 dollars of ETH back on Ethereum from Base. Native route: no bridge fee beyond gas, but the funds are locked for about seven days. Fast bridge charging, say, 0.10%: 2,000 x 0.0010 = 2 dollars plus gas, and the funds arrive in minutes. The fee buys a week of access; the cost is that you now trust the bridge operator's contracts as well as the rollup's.

Always verify that the bridge contract and site are official. Bridge exploits remain a significant attack vector; our guide to what you trust when you cross a bridge explains the models. Send a small test amount first, on the same network on both sides.

Frequently asked questions

How much does a transaction cost on Arbitrum? It varies with demand and with what Ethereum charges for the data Arbitrum posts to it, so check a live fee tracker or your wallet's estimate before you send. Because a rollup bundles many transactions into one mainnet batch, and since March 2024 can post that data as cheaper blobs, fees are usually far lower than on Ethereum itself.

How long does it take to withdraw from an L2 to Ethereum? Through the native bridge of an optimistic rollup, about a week: Arbitrum's challenge period is 6.4 days and OP Stack chains such as OP Mainnet and Base use 7 days. Fast bridges deliver in minutes for a fee, at the cost of trusting the bridge operator.

Does Base have a native governance token? Base launched without one. In September 2025 Base said it was beginning to explore a network token, while Coinbase said it had no definitive plans to launch one. Treat any token marketed as "the Base token" with suspicion until Base announces it on its own channels.

What is the main difference between optimistic and ZK rollups? Optimistic rollups assume transactions are valid and allow them to be challenged with fraud proofs, which creates a withdrawal delay of about a week. ZK rollups prove each batch cryptographically, so correctness does not depend on a challenge window.

What do L2BEAT stages mean? They measure how far a rollup has moved from operator control to code control: Stage 0 is fully controlled by a few entities, Stage 2 is fully controlled by code, and Stage 1 is in between. L2BEAT notes that the stages measure decentralisation, not overall security.

What a Layer 2 Actually Inherits

The security claim for a rollup is precise, and it is narrower than the marketing:

• Data availability — transaction data is published to the base layer, so anyone can

reconstruct the state independently.

• Settlement — the base layer is where disputes are resolved and where the canonical

record lives.

• Not liveness. A layer 2's own sequencer can stop, and most rollups today run a

single sequencer operated by the team.

• Not upgrade immutability. Most have upgrade mechanisms controlled by a multisig or

a security council, which is a real party who can change the rules.

The practical version: rollups inherit the base layer's ability to eventually make you whole, not its resistance to being paused. Escape hatches exist on the better ones and are typically slow and manual.

The Two Architectures, and Where It Matters

The withdrawal delay is the one that affects users directly, and it is why "fast bridge" services exist. Understanding what those are is the next section.

Bridge Risk Is the Largest Risk Here

Bridges have been among the most costly failure points in the entire ecosystem, and the distinction between types is essential:

• The canonical bridge operated by the rollup itself, which uses the security

described above and is slow to exit on optimistic chains.

• Third-party fast bridges, which front you the funds on the destination chain and

take the slow exit themselves. You are trusting that provider, not the rollup.

• Cross-chain bridges between unrelated chains, which hold locked assets and are the

category with the worst historical record.

Practical rules: prefer the canonical bridge unless the delay genuinely matters, test with a small amount first, verify the bridge address from official documentation rather than from a search result, and treat anything wrapped as carrying the risk of whatever holds the original.

Choosing Where to Transact

• Where the application you want actually lives. Liquidity and applications

concentrate, and being on the wrong chain means bridging again.

• How you will get out, decided before you go in.
• Whether the sequencer is decentralised yet, and what happens if it stops.
• Total round-trip cost, including both bridge legs, rather than the per-transaction

fee once you are there.

Capital at risk. Bridging and smart-contract risk apply to every layer 2, and what DeFi is and whether it is safe covers the protocol layer that sits on top.

The bottom line

Layer 2 networks make Ethereum cheaper to use by sharing the cost of mainnet transactions, but each one adds its own operator, bridge and smart-contract risks. Pick the network your app actually runs on, check its L2BEAT stage, bridge only through official routes, and start with a small test amount. For the fee mechanics underneath, read Ethereum gas fees explained, and for how Ethereum compares with other chains, Bitcoin, Ethereum and Solana compared.

Capital at risk. This is not financial advice.