The best Ethereum L2s in 2026 are judged on three numbers that move week to week: median transaction fee, total value locked (TVL), and sustained transactions per second (TPS). As of April 2026, L2BEAT tracks 73 active rollups with combined TVL above $48 billion, and the top eight networks process over 320 transactions per second on a typical weekday. This guide compares those eight on fees, TVL, throughput, security model, and stablecoin liquidity, then explains how the differences map to real builder decisions.
​

The L2 landscape now splits cleanly into two camps. Optimistic rollups (Arbitrum, Base, Optimism, World Chain) dominate TVL and ecosystem maturity. ZK rollups (zkSync Era, Linea, Scroll, Starknet) lead on finality time and proof economics. The gap between camps is closing fast: Arbitrum shipped BOLD permissionless validation in late 2025, and Linea's prover throughput crossed 70 TPS by Q1 2026 per Lineascan. Choosing an L2 is no longer a security versus cost trade — it is an ecosystem and finality decision.
​

An Ethereum Layer 2 (L2) is a separate blockchain that batches user transactions, executes them off the Ethereum mainnet, and posts cryptographic proofs or fraud-proof commitments back to L1 for settlement. The reader gets faster confirmations and fees roughly 10–100x cheaper than L1, while inheriting Ethereum's security guarantees through the proof system.
​

The two dominant proof models are optimistic rollups and zero-knowledge rollups. Optimistic rollups (ORs) assume transactions are valid and allow a 7-day fraud-proof challenge window before withdrawals to L1 finalize. ZK rollups generate validity proofs (zk-SNARKs or zk-STARKs) for every batch, which mainnet verifies in seconds — withdrawals can finalize in under an hour once a proof is posted.
​

L2BEAT's scaling summary classifies each network by stage: Stage 0 (training wheels), Stage 1 (limited council overrides), Stage 2 (fully trustless). As of April 2026, only Arbitrum One and DeGate sit at Stage 1; the rest of the top eight remain at Stage 0, meaning a multisig council can pause or upgrade contracts. The stage rating matters more than marketing claims when picking a chain for treasury-grade flows.
​

Fee comparisons need a fixed reference action. Across this guide, "median fee" means the median cost of an ERC-20 transfer (USDC) over a 7-day window ending April 24, 2026, sourced from growthepie. Swap fees and contract calls run 1.5–4x higher.
​

TVL combines bridged ETH, bridged stablecoins, and native protocol locked value. DefiLlama's chain dashboard separates "canonical bridged" (assets locked on the L2's official bridge) from "third-party bridged" (assets routed via Across, Hyperlane, LayerZero, or others). A network with high canonical TVL signals durable user deposits; high third-party-only TVL signals transient liquidity.
​

TPS measurements vary by methodology. The numbers below use Chainspect's 24-hour rolling average of confirmed transactions, excluding system-generated batch posts. Spikes during NFT mints or token launches can push individual chains past 1,000 TPS for short windows, but those numbers are not sustainable.
​

The following table summarizes the live state of the top eight L2s by TVL as of April 24, 2026. Numbers come from L2BEAT, DefiLlama, Chainspect, and growthepie, cross-checked against block explorers.

Two observations on this snapshot. Arbitrum and Base together hold over half of all L2 TVL, reflecting strong DeFi protocol coverage and large stablecoin treasuries on those chains. ZK rollups sit at lower TVL but post higher fee-per-transaction efficiency once EIP-4844 blob usage is averaged in — Linea spent $0.012 per transaction in blob fees during Q1 2026, versus $0.018 on Arbitrum, per growthepie's blob-cost feed.
​

Arbitrum is the largest L2 by TVL and the only one currently at L2BEAT Stage 1, meaning fraud proofs are permissionless. For a side-by-side with Base, see the Base vs Arbitrum comparison. Its Nitro stack uses interactive fraud proofs and BOLD validation. The ecosystem includes GMX, Aave v3, Pendle, Camelot, Radiant, and Treasure DAO. Native USDC arrived via Circle's CCTP integration in 2023; combined stablecoin float on Arbitrum exceeded $4.2B in April 2026.
​

Base, built on the OP Stack and operated by Coinbase, has grown faster than any L2 in the past 18 months — TVL went from $2.1B in October 2024 to $11.2B by April 2026, per DefiLlama. Base inherits OP Stack's modularity and shares fault-proof code with Optimism. Its largest applications are Aerodrome (the dominant ve(3,3) DEX), Morpho, and a long tail of consumer-facing apps onboarded via Coinbase's distribution. Median fees of $0.02 are the lowest of the top eight.
​

OP Mainnet is the canonical OP Stack chain and the parent of the Optimism Superchain, the parent of the Superchain ecosystem. Stablecoin TVL skews toward USDC and DAI; native protocols include Velodrome, Synthetix, and Sonne Finance. The Superchain initiative has now onboarded Base, World Chain, Mode, Worldcoin, Zora, and several enterprise chains, sharing a sequencer fee pool and unified bridge contracts.
​

zkSync Era from Matter Labs uses zk-SNARK proofs and a Boojum prover upgrade that landed in 2024. Native account abstraction (no separate ERC-4337 layer) is the chain's unique developer angle. TVL is concentrated in SyncSwap, Maverick Protocol, and Aave v3. The 2025 ZK token airdrop drove a usage spike that has since cooled, leaving steady-state TPS around 28.
​

Linea, built by ConsenSys, is the largest zkEVM by TVL after zkSync Era. Its prover system uses gnark and produces proofs every 60–90 seconds on average. Linea's deep MetaMask integration drives high retail wallet activity; the ecosystem leans into LRT/ETH-yield protocols, with Renzo, ether.fi, and EigenPie holding outsized share of TVL.
​

Scroll positions as the most bytecode-equivalent zkEVM, mirroring EVM behavior so closely that mainnet contracts deploy without modification. The trade-off is lower throughput — 14 TPS average — because the prover must handle exact EVM opcodes. Scroll's roadmap calls for 30+ TPS by year-end 2026 once parallelized proving ships.
​

World Chain — also covered in the newer L2s comparison — built on OP Stack, operated by Tools for Humanity, optimizes for World ID-verified humans rather than open-throughput DeFi. Verified users get free gas via priority blockspace, funded by the chain's sequencer revenue. Most TVL is held by the WLD token's onchain treasury and a small ecosystem of payments apps.
​

Starknet uses zk-STARKs (no trusted setup) and a custom Cairo VM. Account abstraction is native at the protocol layer. The ecosystem (JediSwap, Ekubo, Nostra) is smaller than the EVM rollups but has the deepest perps liquidity outside Arbitrum's GMX, with Paradex hitting $400M+ daily perp volume in March 2026.
​

For payments, treasury, and stablecoin orchestration teams, total stablecoin liquidity matters more than total TVL. The full per-chain breakdown lives in the stablecoin liquidity guide. The following ranks the top eight L2s by stablecoin TVL (USDC, USDT, DAI, USDS, FDUSD combined) per DefiLlama on April 24, 2026.

Arbitrum and Base together hold $8.1B in stablecoins — roughly 64% of the top-eight stablecoin float. That concentration matters because CCTP native-USDC routes between mainnet and L2 only exist for chains Circle has integrated; cross-chain teams typically prioritize Arbitrum, Base, OP, and Polygon when picking a deployment order.
​

Every L2 in the top eight today runs a centralized sequencer. The decentralization roadmap differs sharply per chain — covered in the L2 sequencers guide. The sequencer orders transactions, submits batches to L1, and earns priority fees. Centralization here is the largest unresolved trust assumption — a malicious sequencer cannot steal funds (the proof system prevents that), but it can censor transactions or extract MEV.
​

Decentralization roadmaps differ sharply. Arbitrum has shipped a timeboost MEV auction for sequencer slots and signaled a multi-party sequencer for late 2026. Optimism's roadmap calls for shared sequencing across the Superchain, with Espresso Systems as a potential operator. zkSync Era runs a single sequencer today but ships a based-rollup mode in testnet, where Ethereum L1 validators handle ordering. Linea, Base, Scroll, and World Chain have not committed to public timelines.
​

The single-sequencer reality means that for the next 12–18 months, every L2 retains a liveness dependency on the operator. Builders accept this because the alternative — building on a chain with weaker security or worse UX — is not yet competitive.
​

Bridging cost is a frequent surprise for new L2 users. For the full breakdown of native versus third-party paths, see the L2 bridging cost comparison. The cheapest path between two L2s is rarely the canonical bridge. Canonical bridges via L1 cost $4–18 in mainnet gas plus a 7-day delay (for optimistic rollups). Third-party bridges like Across, Hyperlane, and LayerZero route through liquidity pools and complete in under 60 seconds, charging 0.05–0.20% of notional plus a $0.50–$3 fee.
​

For stablecoin transfers specifically, Circle's CCTP (Cross-Chain Transfer Protocol) is the cheapest native option — it burns USDC on the source chain and mints fresh USDC on the destination, with no slippage and roughly $0.30–$0.80 in combined gas. CCTP supports Arbitrum, Base, OP Mainnet, Polygon, Avalanche, and Solana as of April 2026, with Linea and zkSync Era integrations announced.
​

Eco is a stablecoin execution network that handles cross-chain stablecoin movement across all eight L2s above plus seven more chains. Eco Routes (CLI + API) abstracts L2 selection: a developer submits a stablecoin transfer intent, and the network solves for the cheapest, fastest path across canonical bridges, CCTP, and third-party rails. Teams building on multiple L2s use Routes to avoid stitching together a separate bridge integration per chain. Production users include payments providers moving USDC between Arbitrum and Base, treasury teams rebalancing across the Superchain, and trading desks settling cross-VM stablecoin flows.
​

For comparison shoppers picking a primary L2: if your application is stablecoin-heavy and needs deep liquidity, Arbitrum or Base. If you want the lowest finality time for withdrawals back to L1, a ZK rollup. If you need the most decentralized fraud-proof system available today, Arbitrum One. The right answer depends on what you optimize for, not on a single ranking.
​

Base and World Chain tied at $0.02 median fee for a USDC transfer in April 2026, per growthepie. Base wins on ecosystem depth; World Chain offers free gas for World ID-verified users. For ZK rollups, Linea ($0.04) is currently the cheapest. Fees fluctuate with L1 blob prices and can change within a single week.
​

L2 security inherits from Ethereum L1 through the proof system, but security stages vary. Only Arbitrum One sits at L2BEAT Stage 1 (permissionless fraud proofs); the others remain at Stage 0, where a council multisig can pause contracts. Treasury-grade flows should treat Stage 0 chains as needing additional operational diligence.
​

Optimistic rollups assume transactions are valid and use a 7-day fraud-proof window. ZK rollups generate cryptographic validity proofs verified on L1 in seconds. ZK withdrawals to L1 finalize in under an hour; optimistic withdrawals take 7 days unless routed via a third-party bridge that fronts liquidity.
​

Arbitrum One leads with $4.2B in stablecoin TVL as of April 2026, followed by Base at $3.9B. Combined, Arbitrum and Base hold roughly 64% of stablecoin liquidity across the top eight L2s, which is why most cross-chain stablecoin protocols launch there first.
​

If your application needs fast L1 withdrawals (under an hour) or you care about proof-system finality, pick a ZK rollup like zkSync Era, Linea, or Scroll. If you need the largest ecosystem of existing DeFi integrations, deepest liquidity, and most mature fraud-proof system, pick Arbitrum or Base. Most teams that need both deploy to one of each.
​

ZK rollup withdrawals finalize in 1–24 hours once a validity proof is posted to L1. Optimistic rollup canonical withdrawals require a 7-day fraud-proof challenge window. Most users withdraw via third-party liquidity bridges like Across or Hop, which front the funds and complete in under 5 minutes for a 0.05–0.20% fee.
​

Arbitrum launched in August 2021 and accumulated DeFi protocols across two full market cycles, including GMX, Pendle, and Aave v3 deployments. Base launched in August 2023 and grew faster but had less time to build deep protocol coverage. As of April 2026, Arbitrum holds $13.8B versus Base's $11.2B, with the gap shrinking each quarter.
​