State-access gas cost update

Abstract

This EIP updates the gas cost of state-access operations to reflect Ethereum’s larger state and the consequent slowdown of these operations. It raises the base costs for STORAGE_WRITE, COLD_STORAGE_ACCESS, and COLD_ACCOUNT_ACCESS and updates the access cost for EXTCODESIZE and EXTCODECOPY.

Motivation

The gas price of accessing state has not been updated for quite some time. EIP-2929 was included in the Berlin fork in March 2021 and raised the costs of state-accessing opcodes. Yet, since then, Ethereum’s state has grown significantly, thus deteriorating the performance of these operations. This proposal further raises state access costs to better align them with the current performance of state access operations, in relation to other operations.

Additionally, EXTCODESIZE and EXTCODECOPY have the same access cost as BALANCE and EXTCODEHASH. However, EXTCODESIZE and EXTCODECOPY require two database reads - first to load the account object (which includes the nonce, balance, codeHash and storageRoot), and second to collect the required information (the code size and bytecode respectively). Therefore, the access cost of EXTCODESIZE and EXTCODECOPY should be higher when compared with the other account read operations.

Specification

Parameters

Upon activation of this EIP, the following parameters of the gas model are introduced/updated:

¹: 6,700 = CALL_VALUE (9,000) - CALL_STIPEND(2,300)

²: 2,800 = GAS_STORAGE_UPDATE (5,000) - GAS_COLD_SLOAD (2,100) - GAS_WARM_ACCESS (100)

³: 7,000 = GAS_CREATE (32,000) - GAS_NEW_ACCOUNT (25,000)

SSTORE pricing

In addition, the SSTORE cost formula is updated to include the following independent costs:

1. Access costs: COLD_STORAGE_ACCESS or WARM_ACCESS, depending on whether the storage slot is cold or warm.
2. Write cost: additionally charge STORAGE_WRITE if the new value is different from the original value.
3. State creation cost: additionally charge GAS_STORAGE_SET (per EIP-8037) if the original value is zero, the current value is zero, and the new value is non-zero.

Refunds are also updated as follows:

1. STORAGE_CLEAR_REFUND is refunded if the original value is non-zero, the current value is non-zero and the new value is zero.
2. STORAGE_WRITE is refunded if the original value is the same as the new value.

These refunds go to the transaction’s refund counter and are capped to 20% of the total gas used by the transaction, as per the current refund mechanism.

The rules for state creation cost charges and refills are addressed in EIP-8037.

Cases and their corresponding costs are detailed in the table below:

Account access pricing

The rules that determine when COLD_ACCOUNT_ACCESS and WARM_ACCESS are charged for account-related operations are unchanged; only the parameter values themselves are updated.

CALL/CALLCODE

For CALL and CALLCODE operations, CALL_VALUE is set to ACCOUNT_WRITE + CALL_STIPEND, where CALL_STIPEND = 2,300.

CREATE/CREATE2

CREATE and CREATE2 operations don’t have explicit warm/cold pricing. Instead, they are charged CREATE_ACCESS in regular-gas and GAS_NEW_ACCOUNT in state-gas (per EIP-8037). CREATE_ACCESS is defined as ACCOUNT_WRITE + COLD_STORAGE_ACCESS.

Note: this definition does not exactly match the legacy decomposition shown in footnote ³ of the Parameters table (GAS_CREATE - GAS_NEW_ACCOUNT = 7,000). The pre-existing accounting in the protocol is not internally consistent here, and this EIP keeps the discrepancy rather than attempting to reconcile it.

SELFDESTRUCT

For SELFDESTRUCT, an additional charge of ACCOUNT_WRITE is added if a positive balance is sent to an empty account.

EXT* family update

Besides the current access costs, EXTCODESIZE and EXTCODECOPY are charged an additional WARM_ACCESS.

EIP-7702 authorizations pricing

EIP-7702 authorizations have two separate gas costs - intrinsic costs and access costs. Intrinsic costs are computed during transaction validation, while access costs are charged during code execution. The following table summarizes these costs:

REGULAR_PER_AUTH_BASE_COST is defined in EIP-8037 as the sum of:

• Calldata cost: 1,616 (101 bytes × 16)
• Recovering authority address (ecRecover): PRECOMPILE_ECRECOVER (updated by EIP-7904)
• Reading nonce and code of authority (cold access): COLD_ACCOUNT_ACCESS (updated by this EIP)
• Storing values in already warm account: 2 x WARM_ACCESS

Rationale

Empirical Estimation of Gas Costs

Selecting which Operations to Reprice

The parameters selected for repricing in this EIP were chosen based on the estimated million gas per second (Mgas/s) performance of their respective operations. For this EIP, a performance target of 100 million gas per second is chosen. With this performance target, we will be able to increase the base throughput of the chain by 5x from our current 20Mgas/s performance.

Data Collection

The gas costs proposed in this EIP are based on the actual time each execution client takes to run a specific set of benchmarks. Similar to the methodology used in the Gas Cost Estimator project, we generate synthetic blocks that isolate and stress individual EVM operations and use them to derive the various gas parameters.

Concretely, to benchmark a single operation/parameter, different blocks are created by varying the number of times the target operation is executed and by changing the parameter values to the operation. The EEST benchmark suite contains the tests used to generate these blocks.

Then, the Benchmarkoor tool is used to collect the needed metrics. This tool returns the total execution time of the block and the number of times each operation was executed. Each test block is run multiple times on each client to account for variability in execution time.

All benchmarks were run on top of a snapshot with the same size as mainnet in March 2026, so they reflect the current performance of state access operations.

What about differences between client implementations?

Ultimately, we require a single cost model that is independent of the client’s implementation. Although the decision may be different in a specific situation, the general rule is to take the worst client result that cannot be optimized. This means that the worst-performing client on a given operation and resource combination will define the cost of that operation for the entire network. This is the safer and most conservative choice.

Runtime Estimation

To estimate the runtime of a single operation from the total execution time of the corresponding synthetic blocks, a Non-Negative Least Squares (NNLS) Linear Regression is used. This model enforces that all coefficients are non-negative, thus ensuring execution time cannot be negative. The model estimates runtime as a linear combination of:

• Constant term (intercept): Base overhead for executing the test, which includes setup and teardown time
• Operation count (slope): Number of times the operation is executed in the test. This parameter is the one that allows us to estimate the per-operation runtime.
• Operation-specific parameters (e.g., update): Extra per-operation runtime contributed by an additional, operation-specific binary or numeric input (e.g., whether a CALL is made with value, or whether the SSTORE writes a new value). The update coefficient gives the marginal runtime added per execution for storage writes.

For simple operations (i.e., without additional parameters), the model estimates: runtime = intercept + slope × operation_count

For variable operations, the model estimates: runtime = intercept + slope × operation_count + param1_coef × operation_count × param1 + param2_coef × operation_count × param2 + ...

Independent models are created for each client, operations, and configuration. Only operations and parameters with good model fits (R² > 0.5 and p-value < 0.05) are included in the gas cost proposal to ensure the reliability of the estimates.

Glue opcode adjustment

Each benchmark test includes auxiliary “glue” opcodes that scale linearly with the main opcode count. The regression slope captures the combined runtime of both the target opcode and its glue opcodes, so we subtract the estimated glue opcode runtime to isolate the true per-execution cost:

adjusted_slope = slope - sum(ratio_i * glue_runtime_i)

Where, for each glue opcode i:

• ratio_i is the number of executions of glue opcode i per execution of the target opcode in the benchmark test.
• glue_runtime_i is the estimated per-execution runtime of glue opcode i, taken from its own regression model.

Only glue opcodes with a statistically significant fit (p-value < 0.05) are included. The adjusted slope is clipped to a minimum of zero.

Parameter mapping

Each gas parameter is derived from one or more benchmark models by selecting a specific regression coefficient (slope or update) after applying the relevant filters (test name, cache strategy, account mode).

For each parameter and client, the worst-case value across all matching model configurations is used. The final proposal takes the worst case across clients.

The following table lists the directly estimated parameters and their mapping to the regression models:

test_sstore_bloated and test_sload_bloated targeted a bespoke contract with a storage size of 10GB, which is the size of the largest mainnet contract.

Conversion to gas costs

New gas costs are calculated using the same target performance of 100Mgas/s. The formula used is:

new_gas = (anchor_rate * runtime_ms) / 1000

Where:

• anchor_rate is the target performance expressed in gas per second (e.g., 100_000_000 for a 100Mgas/s target).
• runtime_ms is the estimated runtime per operation, in milliseconds, taken from the regression models (the adjusted_slope defined above).
• The factor 1000 converts runtime_ms from milliseconds to seconds, so that the product with anchor_rate (gas/second) yields a result in gas units.

Derived parameters are computed from the directly estimated ones:

STORAGE_CLEAR_REFUND         = (STORAGE_WRITE + COLD_STORAGE_ACCESS) * (4800/5000)
ACCESS_LIST_STORAGE_KEY_COST = COLD_STORAGE_ACCESS
ACCESS_LIST_ADDRESS_COST     = COLD_ACCOUNT_ACCESS
CREATE_ACCESS                = ACCOUNT_WRITE + COLD_STORAGE_ACCESS

Interaction with EIP-7928

EIP-7928 introduces Block-Level Access Lists, which enable parallel disk reads, parallel transaction validation, and executionless state updates through client optimizations. Our benchmarks already consider these optimization gains by running on clients with the optimizations implemented.

Special case for EXTCODESIZE and EXTCODECOPY

Differently from other account read operations, EXTCODESIZE and EXTCODECOPY make two reads to the database. The first read is the same, where the object of the target account is loaded. This object includes the account’s nonce, balance, codeHash and storageRoot. Then, these operations do another read to collect either the code size or the bytecode of the account. This second read is to an already warmed account, and thus we propose to price it as WARM_ACCESS for consistency.

Interaction with EIP-8032

EIP-8032 proposes modifying the SSTORE cost formula to account for the storage size of a contract. This is a more accurate method for pricing storage writes. It allows writes to smaller contracts to be cheaper than writes to large contracts, which helps with scaling and is fairer to users. However, EIP-8032 is not yet scheduled for inclusion in a fork, and, as such, this proposal considers two cases for setting the values of STORAGE_WRITE and COLD_STORAGE_ACCESS:

1. Before EIP-8032. In this case, we set the parameters assuming a worst-case contract size, which makes state-accessing operations more expensive, independently of the contract size.
2. After EIP-8032. In this case, we set the parameters assuming the worst-case until ACTIVATION_THRESHOLD, which is the parameter in EIP-8032 that triggers the depth-based cost. This means that contract sizes below the threshold have the same access costs, while contracts above the threshold get exponentially more expensive with increasing size.

Interaction with EIP-2926

EIP-2926 proposes to change how code is stored in the state trie. One of the changes of this proposal is to include the code size as a new field in the account object (codeSize). With this change, the code size can be directly accessed with a single read to the database and thus EXTCODESIZE should maintain its previous cost formula, without including the additional WARM_ACCESS.

Backwards Compatibility

This is a backwards-incompatible gas repricing that requires a scheduled network upgrade.

Wallet developers and node operators MUST update gas estimation handling to accommodate the new state access cost rules. Specifically:

• Wallets: Wallets using eth_estimateGas MUST be updated to ensure that they correctly account for the updated gas parameters. Failure to do so could result in underestimating gas, leading to failed transactions.
• Node Software: RPC methods such as eth_estimateGas MUST incorporate the updated formula for gas calculation with the new state access cost values.

Users can maintain their usual workflows without modification, as wallet and RPC updates will handle these changes.

Security Considerations

Changing the cost of state access operations could impact the usability of certain applications. More analysis is needed to understand the potential effects on various dApps and user behaviors.

Copyright

Copyright and related rights waived via CC0.