Account abstraction on Ethereum and its layer-2s now ships in three distinct flavors: ERC-4337 (a smart-contract wallet standard riding an alternate mempool), ERC-7702 (a consensus-level delegation that lets an externally owned account temporarily execute smart-contract code), and native AA (zkSync Era and Starknet, where every account is a contract at the protocol level). Each approach answers the same question, "how does a user account get programmable behavior," with different tradeoffs in gas cost, infrastructure dependency, security model, and migration path. This article compares the three side by side and shows when each is the right choice in 2026.
​

Account abstraction breaks Ethereum's split between externally owned accounts (EOAs, controlled by a private key) and contract accounts (controlled by code). Under AA, any account can carry custom validation logic such as multisig rules, social recovery, session keys, gas sponsorship, and batched calls. The three live standards in 2026 reach this goal at different layers of the stack.
​

The split has been a known limitation since 2015. EOAs always pay their own gas in ETH, sign with one private key, and execute one call per transaction. Contract wallets like Safe (formerly Gnosis Safe) and Argent have offered programmable rules since 2018, but funding them required a counter-party relayer or a pre-funded balance. ERC-4337, ERC-7702, and native AA each remove a different friction point in that pattern.
​

For background on the broader pattern, see the ERC-4337 explainer and the account abstraction stack overview. The original motivation is documented in Vitalik Buterin's EIP-4337 forum post and his 2023 essay "The Three Transitions," which names account abstraction as one of three required upgrades for Ethereum to reach mainstream usability.
​

ERC-4337 introduces a parallel transaction format called a UserOperation that flows through an alternate mempool. Bundlers collect UserOps, validate them against a singleton EntryPoint contract, and submit them as one regular Ethereum transaction. No consensus change is required, which is why the standard reached final status on the Ethereum mainnet in March 2023 without a hard fork.
​

The mechanism has four roles. A smart-contract wallet implements the validation logic. A bundler is a permissionless node that aggregates UserOps from the alternate mempool. The EntryPoint contract verifies signatures, debits gas, and dispatches the call. A paymaster, when present, sponsors gas in ETH or stablecoins so the wallet itself never has to hold ETH. The reference EntryPoint v0.7 is deployed at 0x0000000071727De22E5E9d8BAf0edAc6f37da032 on every major EVM chain and is the singleton that wallets and bundlers agree on.
​

The standard is documented in EIP-4337 and the eth-infinitism reference implementation. The BundleBear dashboard tracks UserOp volume across chains. As of Q1 2026, Base and Polygon carry the largest share of weekly UserOps, with Coinbase Smart Wallet and Pimlico-served wallets accounting for most of that volume.
​

ERC-7702 is a consensus-level upgrade that lets an EOA delegate its execution to a smart-contract address. The EOA keeps its existing address, balance, and nonce, while gaining the ability to run code such as batched calls, paymaster-funded gas, or session-key checks without migrating to a new account. The standard shipped in the Pectra hard fork on May 7, 2025.
​

The mechanism adds a new transaction type, 0x04, that carries an authorization list. Each entry binds a target contract address to the EOA's key with a signature. While the transaction executes, the EVM treats the EOA's code as if it were the delegated contract's code. After execution, the delegation persists onchain until the EOA revokes or rebinds it. This means a Metamask or Rabby user can install a smart-wallet implementation onto their existing address without moving any funds.
​

The text of the standard is at EIP-7702 and the Pectra upgrade is documented at ethereum.org/roadmap/pectra. ERC-7702 replaced the older EIP-3074 proposal, which was deprecated in early 2024 in favor of this approach. Vitalik's May 2024 post summarizes the rationale for the swap, citing 7702's simpler interaction with the rest of the AA stack.
​

Native account abstraction collapses the EOA versus contract distinction at the protocol level. On zkSync Era and Starknet, every account is a contract from the first byte of bytecode. There is no EntryPoint, no alternate mempool, and no separate UserOp transaction type. The chain's protocol natively understands custom validation logic, paymasters, and sponsored transactions as first-class primitives.
​

On zkSync Era, the system contract ContractDeployer creates account contracts using the same code path as any other contract, and the BootloaderUtilities system contract validates signatures and dispatches calls. zkSync's native AA shipped on mainnet in March 2023, predating ERC-4337's final adoption. On Starknet, every account derives from the OpenZeppelin Account or a custom Cairo contract from genesis. There has never been an EOA on Starknet.
​

Documentation lives at zkSync's AA reference and Starknet's accounts documentation. The Argent X for Starknet wallet is the most-used native AA wallet by transaction count, per Starknet block explorer data. Native AA chains do not require ERC-4337 infrastructure because the chain itself plays the EntryPoint role.
​

The three approaches reach overlapping outcomes through different mechanisms, with measurably different gas costs and infrastructure dependencies. ERC-4337 layers on any EVM chain without protocol changes, ERC-7702 lets existing EOAs opt into smart-wallet behavior after the Pectra fork, and native AA bakes the model into chain genesis. The comparison table below summarizes the tradeoffs across six dimensions.
​

Gas figures track measurements from BundleBear, Dune's ERC-4337 dashboard, and the eth-infinitism gas analysis. The 30 to 80 percent ERC-4337 overhead comes from EntryPoint validation, bundler signature checks, and the multiple SLOAD operations needed to verify a UserOp. ERC-7702's overhead is lower because the validation logic runs on a delegated contract that the EVM already loads in the normal transaction flow. Native AA's overhead is the smallest because the chain's prover or sequencer treats account validation as part of base execution rather than a layered protocol.
​

The three standards are not mutually exclusive. An ERC-7702 delegation can point at a smart-wallet implementation that also speaks the ERC-4337 UserOp format, letting that account use ERC-4337 paymasters and bundlers when the user wants gas sponsorship. The MetaMask Smart Account, released in late 2025, uses exactly this pattern. Native AA chains can also expose ERC-4337-compatible interfaces for tooling compatibility, though they do not need them for users.
​

The choice depends on which EVM chain the application targets, what wallet population it serves, and how much infrastructure the team is willing to operate. ERC-4337 fits brand-new wallets and dapps that need full smart-wallet features across many chains. ERC-7702 fits applications upgrading existing EOA users without forcing a migration. Native AA fits teams committed to a single L2 ecosystem.
​

The decision tree maps onto three common product cases. A new consumer wallet shipping a self-custodial product across Ethereum mainnet and the leading L2s typically picks ERC-4337 because it works everywhere with the same code. Coinbase Smart Wallet, launched in June 2024 as a passkey-based ERC-4337 wallet, follows this path. A wallet vendor with an installed EOA base, like Metamask, leans on ERC-7702 because it can ship smart-account features to existing users without an address migration. A team building on zkSync Era or Starknet uses native AA because no other option exists and the chain's tooling assumes it.
​

Three secondary factors push the decision. Gas-sponsorship economics favor ERC-7702 for high-frequency apps because the overhead is lower; see the paymaster article for sponsorship details. Cross-chain transferability of a wallet's address matters for self-custody products targeting multiple ecosystems; ERC-4337 wallets can use the same factory address across chains, while native AA accounts cannot exist outside their home chain. Audit cost matters for teams using custom validation logic; ERC-4337 leans on audited reference contracts (Safe, Kernel, Biconomy Nexus), ERC-7702 inherits the security of whatever contract the EOA delegates to, and native AA chains audit account behavior at the protocol layer.
​

Each standard concentrates risk in a different place. ERC-4337 puts the security boundary at the EntryPoint contract, bundler mempool rules, and the wallet's validation code. ERC-7702 places it on the contract the EOA delegates to, which has full control of the address. Native AA pushes it into the chain's protocol code, where account validation is part of consensus.
​

The ERC-4337 EntryPoint at v0.7 has been audited by OpenZeppelin and Code4rena, with audit reports linked from the eth-infinitism audits directory. Bundler mempool rules ban opcodes that could let a malicious UserOp grief the bundler, listed in ERC-7562. Smart-wallet implementations like Safe and Kernel have their own audit histories spanning multiple years.
​

ERC-7702's main novel risk is signature replay across chains, addressed in the EIP by binding the authorization to a chain ID and nonce. The wallet client must still warn users before they sign a delegation to an unfamiliar contract, since a malicious delegation target can drain the address. Native AA chains move the audit surface into the protocol; zkSync's bootloader code is open source and has been audited by OpenZeppelin and Halborn, while Starknet's account validation is enforced by the Cairo prover.
​

Pectra's May 2025 inclusion of ERC-7702 closed the question of whether Ethereum L1 would adopt consensus-level AA. The next AA-relevant upgrade is Fusaka, scheduled for late 2026, which is expected to include further EVM Object Format (EOF) changes that improve smart-wallet gas costs. Native AA on Ethereum L1 itself is not on the near roadmap; ERC-7702 is the L1 path.
​

EIP-3074, the prior consensus-level AA proposal that introduced AUTH and AUTHCALL opcodes, was withdrawn in early 2024 in favor of ERC-7702. The Ethereum Magicians thread at "EIP-3074 vs EIP-7702" records the discussion. The L2 picture is fragmented: Optimism and Arbitrum carry ERC-4337 plus ERC-7702 since their respective post-Pectra hard forks, Base inherits both via the OP Stack, zkSync Era continues with native AA plus optional ERC-4337 interfaces, and Starknet stays native.
​

Vitalik Buterin's May 2024 post on decentralization describes the long-term goal as full account abstraction at L1, with ERC-7702 as the first installment and further EVM changes carrying the rest. The Ethereum roadmap page tracks the public state of these proposals.
​

Eco Routes moves USDC, USDT, and other stablecoins across 15 EVM chains using intent-based execution. Smart-wallet users on any of the three AA models can submit a route as one signed intent, with destination-chain settlement handled by a solver. Gas on either side can be sponsored by the app or paid in the transferred stablecoin.
​

Eco's integration is wallet-agnostic. An ERC-4337 wallet using a Pimlico paymaster can pay Eco Routes gas in USDC. An ERC-7702-delegated EOA on Base can batch a token approval and a route submission in one transaction. A zkSync Era account uses the chain's native paymaster to do the same. The user-facing flow is identical across the three standards because Eco Routes operates above the wallet layer. Details are at the eco.com product page.
​

No. ERC-7702 and ERC-4337 are complementary. ERC-7702 lets an existing EOA execute smart-contract code, and that code can be an ERC-4337-compatible wallet implementation. Combining the two gives an EOA holder access to ERC-4337 bundlers and paymasters without migrating to a new address. Most modern wallets in 2026 support both.
​

Native AA chains do not need ERC-4337 for users, but they can expose ERC-4337-compatible interfaces for tooling. zkSync Era has an experimental EntryPoint deployment for SDK compatibility. Starknet does not, because its account format is incompatible with EVM-style UserOps. Application code targeting a single native AA chain usually uses the chain's native primitives.
​

Ethereum L1 is not adopting full native AA in 2026 or 2027. ERC-7702 is the consensus-level path, and ERC-4337 continues as the optional smart-wallet layer above it. Native AA at L1 would require deeper EVM changes that are not on the public roadmap. ERC-4337 infrastructure will continue to operate for the foreseeable future, including on L2s that already ship ERC-7702.
​

Native AA on zkSync Era and Starknet has the lowest measurable gas overhead because account validation is part of the chain's base execution. ERC-7702 ranks second because the delegated contract runs in the normal EVM transaction flow. ERC-4337 has the highest overhead due to EntryPoint validation, alternate-mempool checks, and the additional storage reads the standard requires.
​

Sources and methodology. EIP texts and dates pulled from eips.ethereum.org. Gas measurements drawn from BundleBear and the eth-infinitism repo as of Q1 2026. Pectra hard-fork date verified against the Ethereum roadmap page. Native AA launch dates verified against zkSync and Starknet documentation.
​

To see how Eco Routes works with ERC-4337, ERC-7702, and native AA wallets across 15 EVM chains, visit eco.com.
​