erc4626-bench

Per-transaction gas benchmarks for ERC4626 vaults, on the EVM and on EraVM, same calls, same script.

The benchmark repos people cite compare three libraries and have not moved since 2024. This one measures every call the way a user pays for it, from an isolated transaction, and keeps the baselines in the tree so anyone can add a vault and get a row.

---

Why per transaction

forge test --gas-report aggregates calls from different tests into one row, so a cold first deposit and a warm tenth deposit end up as "min" and "max" of the same function, and the labels get swapped. script/bench-evm.py runs each benchmark on its own under forge test --isolate, so every call starts cold like a real transaction, and reads the gas of that one call from its trace. The number excludes the 21,000 intrinsic cost, which is the same for every vault.

On zkSync Era, forge's gas report is dominated by a fixed bootloader cost and gasleft() inside the emulator does not measure EraVM execution. script/bench-eravm.sh deploys every vault to a local anvil-zksync node, sends the same transaction sequence to each, and reads gasUsed from the receipts.

Baselines

Vault	What it is
SoladyVault	Solady's ERC4626, the accepted cheapest general-purpose implementation
YulSafeERC20	YulSafe: packed totals, inline assembly on the hot path, first-deposit burn
PlainPackedVault	YulSafe rewritten in plain Solidity with the same layout and zero assembly
LeanVault	Packed totals and pause flag, transient reentrancy guard, single supply counter, first-deposit burn
LeanVault2	LeanVault with a virtual-share offset instead of the burn and a one-slot owner
LeanVault3	leanvault's idle vault as shipped, copied from its source: LeanVault2's accounting behind three virtual hooks, plus EIP-2612 permit

The first five pass a16z's 26 ERC4626 properties in the YulSafe repo, LeanVault3 passes them and its permit tests in leanvault. Only the benchmark itself lives here. Solady and LeanVault3 are the two with permit.

Results

Per transaction, solc 0.8.37, cancun, optimizer at 10,000,000 runs, forge 1.3.5 (foundry-zksync 0.1.4), 2026-09-26; Lean3 added 2026-09-27 and refreshed 2026-09-28 for leanvault's pause change, on the same toolchain, with the other columns reproduced unchanged. CI runs Foundry 1.8.3 and reports YulSafe and Plain 2,800 higher on subsequent deposit, mint, withdraw and redeem; the other four columns are identical there:

Call	YulSafe	Plain	Lean	Lean2	Lean3	Solady
deposit() first, cold vault	175,513	173,805	129,500	106,096	106,096	106,009
deposit() subsequent	63,028	61,551	54,469	54,796	54,796	54,709
mint()	63,078	61,740	54,636	54,843	54,877	54,735
withdraw()	61,279	59,938	52,986	53,264	53,265	54,576
redeem()	61,303	59,818	52,866	53,126	53,122	53,284
totalAssets()	2,321	2,321	2,321	2,321	2,321	5,621
convertToShares()	2,674	2,679	2,656	3,002	3,002	8,072
convertToAssets()	2,675	2,680	2,680	3,004	3,104	8,108
permit()					73,903	76,296
Deployment gas	1,712,163	2,154,694	1,983,630	1,764,437	1,959,733	1,185,598

Lean3 is Lean2 with hooks, permit, and a pause that stops deposits but never an exit. Against Lean2 it costs 34 gas more on mint, 100 more on convertToAssets, 1 more on withdraw, 4 less on redeem, and 195,308 more to deploy; a signed permit runs 2,393 cheaper than Solady's. Deployment gas depends on the source path through the metadata hash: Lean and Lean2 were measured from YulSafe's copies under test/mocks/ and deploy 12 gas cheaper from this repo, Lean3 is measured here.

EraVM, from receipts on anvil-zksync 0.6.11 with zksolc 1.5.15 and the patched solc 0.8.30:

Call	YulSafe	Plain	Lean	Lean2	Lean3	Solady
deposit() first	201,031	198,979	179,832	170,532	171,088	173,466
deposit() subsequent	178,774	176,740	163,369	163,172	163,728	166,106
mint()	178,768	176,734	163,236	162,932	163,488	166,148
withdraw()	176,294	174,272	160,806	160,502	161,052	167,356
redeem()	176,288	174,266	160,800	160,484	161,034	164,200

On EraVM, Lean3 pays 550 to 556 gas per call over Lean2 and still beats Solady on every row; Lean beats it on everything but the first deposit. EraVM charges pubdata by how well each written value compresses, so a call's gas depends on the balances and nonces written before it. Lean3 runs from its own account for that reason: sharing the first account moved Lean's subsequent deposit by 119 gas. permit is not measured on EraVM.

Run it

git clone https://github.com/yodablocks/erc4626-bench && cd erc4626-bench
forge test                                 # the benchmark tests, all vaults
script/bench-evm.py                        # per-transaction table, default profile
FOUNDRY_PROFILE=viair script/bench-evm.py  # same, via-IR at 200 runs

anvil-zksync --port 8011 &                 # needs foundry-zksync
script/bench-eravm.sh

Add a vault

1. Put the contract under src/vaults/ with a constructor (address asset, string name, string symbol).
2. Add it to test/GasBenchmark.t.sol by copying one vault's eight test_gas_* functions.
3. Add its name to gas_reports in foundry.toml, to the vaults list in script/bench-evm.py, and as a column in script/bench-eravm.sh, sending its transactions from an account of its own the way LeanVault3 does, so the existing columns do not move.

License

MIT. Solady and forge-std are vendored under their own licenses in lib/.