Cabinet
Back to Technology Insights
2026.08.20

Glamsterdam Forks on the Platåberget Public Testnet — EIP-8037 Breaks the "Transfers Cost 21,000 Gas" Assumption

Glamsterdam Forks on the Platåberget Public Testnet — EIP-8037 Breaks the "Transfers Cost 21,000 Gas" Assumption

What happened

Platåberget reached genesis on 2026-08-13, was announced by the Ethereum Foundation's Protocol DevOps team on 2026-08-17, and forked to Glamsterdam on 2026-08-20. Its Chain ID is 7091047534. The validator set is permissionless: anyone can deposit test validators through the deposit contract.

What separates this network from earlier devnets is that it did not start as a new chain. Rather than launching from its own genesis state like Sepolia or Hoodi, it starts from a state identical to mainnet and transitions to Glamsterdam rules at the 20 August fork. Contracts currently deployed in production can therefore be observed directly across the fork boundary.

Glamsterdam's meta EIP, EIP-7773, is still Draft, and lists ten proposals as Scheduled for Inclusion. The consensus-layer headliner is EIP-7732 (Enshrined Proposer-Builder Separation, ePBS); the execution-layer headliner is EIP-7928 (Block-Level Access Lists). On top of those sit EIP-8037 (State Creation Gas Cost Increase), EIP-7778 (block gas accounting without refunds), EIP-7708 (ETH transfers emit a log) and EIP-7954 (maximum contract size raised from roughly 24 KiB to 32 KiB). Mainnet is targeted for Q4 2026, having slipped twice from the original first-half-of-2026 goal.

Why state-creation gas had to be separated

EIP-8037 is where the design argument lives. It takes operations that grow state — storage allocation, code deployment, account creation — and charges them outside ordinary execution gas, as a number of state bytes multiplied by a cost_per_state_byte (CPSB). For Glamsterdam, CPSB is fixed at 1530.

Why split them at all? Raising the gas limit increases how much the network can process, but it also increases how fast the state database grows. While both live in the same gas unit, any decision to raise throughput is simultaneously a decision to raise node operators' disk burden. EIP-8037 gives state growth its own budget and decouples the two. Glamsterdam can credibly aim at a 200M gas limit because that ceiling was settled first. This is not a proposal that makes Ethereum faster; it is a proposal that draws the line beyond which going faster would break something.

The devnet-7 test releases go further: under the revised EIP-2780 and EIP-8037, state-dependent transaction costs move out of intrinsic gas and into runtime charges at the top frame, and account-creation state gas is charged conditionally at access. The old pattern of over-charging up front and refunding later is replaced by charging when the state is actually touched.

Where compatibility breaks

The Foundation's DevOps team has said plainly that tools treating gas limits as fixed values will stop working, and has urged immediate updates. This is the change with the widest blast radius.

Concretely: code that hardcodes 21,000 for a transfer, batch jobs that cache and reuse gas estimates, and indexers or wallets that keep gasLimit in a constant table. A transfer to a new account will cost more than a transfer to an existing one. Once the same transfer costs differently depending on whether the recipient already exists, any implementation that does not re-estimate before sending will fail. ePBS and BALs mostly concern validators and builders; gas repricing touches every wallet, indexer and gas estimator on the network.

The contract-size increase in EIP-7954 cuts the other way — contracts previously forced into splits can ship as one. But deployment pipelines and verifiers that validate size against a 24 KiB assumption will produce wrong verdicts until they learn the limit moved.

EIP-7708 makes ETH transfers and burns emit a log. For anyone monitoring, this is welcome: wallets and exchanges that have been running custom transaction tracing to follow internal transfers can get the same information from ordinary log subscriptions. There is a reverse effect, though. The same movement of funds becomes observable in two representations, via traces and via logs, so accounting that naively aggregates both will double-count. Which one is canonical needs deciding before the transition, not during it.

EIP-7778 moves to block gas accounting without refunds. Gas refunds — returning consumed gas when storage is released — have made per-block consumption hard to predict. Removing them improves builder predictability, but contracts that trimmed their gas limits on the assumption of a refund, particularly batch jobs that release large amounts of storage, will see effective costs rise. Any implementation that batches storage clears as a gas optimisation loses the premise of that optimisation.

Pitfalls to watch

First, Platåberget is a testnet forked from mainnet state, not a replacement for Sepolia or Hoodi. Passing there does not license skipping public testnet validation. Conversely, for regression-testing already-deployed contracts, its conditions are closer to production than Sepolia's. The two serve different purposes; plan to run both.

Second, the EIP-8282 builder contracts (deposit and exit) are not part of genesis — they are deployed on-chain before the Gloas fork activates. Anyone testing builder registration needs to confirm when those two addresses go live before writing the procedure.

Third, the changes take effect at the fork height, not at the moment a client is updated. Upgrading the client early changes nothing, which makes it easy to mistake "updated, nothing happened" for a failure. The flip side is that everything switches at once — there is no gradual migration path.

Fourth, reporting around ACDC #185 (20 August) indicates that several clients stalled or slowed immediately after new builder functionality was activated on Platåberget, which is also used internally as Glamsterdam devnet-8; the network was reported stable before activation. That means public testnet dates can move. Pinning internal deadlines to "the Sepolia fork date" lets them slip whenever that date does. Set your own deadline instead.

[Technical Insights & Actions]

  1. Gas cost stops being a single price for execution and becomes two-dimensional: execution plus state growth. Every layer that has treated gas as one scalar — estimation, accounting, limit checks, fee pass-through logic — is affected. This is not Ethereum-specific; it recurs wherever an EVM-compatible chain adopts the same repricing. [Contract developers / integrators] By two weeks before the Glamsterdam fork on Sepolia, audit the codebase for every 21000, every fixed gasLimit and every cached gas estimate, and replace them with runtime estimation.
  2. Platåberget is a public network forked from mainnet state, which makes it a rare chance to measure how currently deployed production contracts behave after the fork. A testnet with its own genesis will not reproduce state-dependent bugs. [Product and QA teams] Before the non-finality devnet begins, run a full regression pass of production-deployed core contracts on Platåberget and record the pre- and post-fork differences.
  3. ePBS turns relays from mandatory into optional, but immediately after the transition both the old and new paths coexist. Monitoring and alerting built around a single path will no longer distinguish which one failed. [Node operators / staking providers] Before the mainnet date is confirmed, exercise block proposal, builder registration and exit end-to-end under an ePBS configuration on testnet, and extend monitoring to cover both paths.

[Sources]

This article is a technical commentary based on publicly available information and is not investment advice.