Ethereum Developers Push Glamsterdam's Sepolia Fork Toward October After Devnet Bugs Hit Three Clients
Core developers on the Sept 3 consensus call discussed a builder-deposit caching bug and non-finality at the Glamsterdam activation on the latest devnet, affecting Lighthouse, Prysm and Teku. A post-call recap puts the Sepolia fork at Oct 6, later than the Sept 28 slot proposed in August.
Ethereum's Glamsterdam upgrade is taking longer to reach public testnets than developers hoped in August. On the September 3 All Core Devs Consensus call (ACDC #186), client teams worked through a postmortem of a builder-deposit caching bug on devnet-8 and a period of non-finality at the Gloas (consensus-layer) activation that affected the Lighthouse, Prysm and Teku clients, according to the call agenda on the ethereum/pm GitHub repository.
Where the schedule stands
The pre-call agenda still carried the August proposal to fork Sepolia at epoch 351,232, which corresponds to September 28. Christine D. Kim's recap of the call, published September 4, reports that developers instead set the Sepolia Glamsterdam fork for October 6 at 13:53 UTC, while noting that no stable devnets were available at the time of the decision. That October date has not yet been confirmed in an Ethereum Foundation post or published call notes, so it should be treated as provisional until it is.
The August plan, as reported by Cryptobenelux after ACDC #185 on August 20, was Sepolia on September 28 followed by Hoodi on October 26. Those calls also surfaced the client-side pain: Lodestar cut its processing time from about 20 seconds to roughly 500 milliseconds, Lighthouse failed on a pre-fork caching optimization, and Teku ran into prolonged data-availability checks. A ninth devnet focused on non-finality testing was awaiting smoke tests as of September 3.
Why Glamsterdam is hard
Glamsterdam is the first fork to ship enshrined proposer-builder separation (ePBS), which restructures how blocks are built and attested, alongside block-level access lists (EIP-7928) and the EIP-8037 gas repricing. That repricing makes new-account creation roughly 8.5 times and contract deployment roughly 10 times more expensive, raises the contract size limit from 24 to 64 KiB and the initcode limit from 48 to 128 KiB, and targets a gas limit around 200 million. The Ethereum Foundation's DevOps team warned in August that any tool relying on a hardcoded maximum gas limit will break. Parithosh Jayanthi of the Foundation has called Glamsterdam probably the largest fork since the Merge.
The Platåberget devnet, which forked on August 20 with about 50,000 validators across roughly 50 nodes, was the first large-scale rehearsal. The bugs it exposed are exactly what devnets exist to find, but each one pushes the testnet and mainnet timeline. Other items on the September 3 agenda included deferred EIP-8037 state-gas questions, the Engine API's migration to SSZ encoding, and a mid-September deadline for ranking EIPs for the following fork, Hegotá.
No mainnet date has been confirmed in any primary source we reviewed. Third-party schedule aggregators show a range of dates, including some earlier than the developers' own discussion, and should not be relied on.