Ethereum will activate the Glamsterdam upgrade on the Sepolia testnet at 13:53:36 UTC on October 6, epoch 353,024, slot 11,296,768, the last scheduled public rehearsal before a mainnet activation the Ethereum Foundation expects in the fourth quarter of 2026. The fork pairs the execution-layer Amsterdam changes with the consensus-layer Gloas changes in one upgrade, and it carries two structural shifts client teams have spent months testing: protocol-level proposer-builder separation and block-level access lists. Hoodi and mainnet dates have not been set, so Sepolia is where the last audit happens in public.
The headline number is gas. Sepolia validators will be able to propose blocks with a limit of 200 million gas, roughly triple what the testnet ran before the fork. Mainnet sits at 45 million after a January increase, so the testnet configuration points at a production ceiling more than four times the level validators approved at the start of the year. The opt-in design matters: Prysm 7.2.0 and Teku 26.9.1 both default to 60 million after the fork, and proposers who want the higher limit have to set it explicitly, through Prysm’s version 2 proposer settings or Teku’s builder registration flag.
ePBS moves block building into the protocol
The deeper change is ePBS, defined in EIP-7732. Today the relationship between block proposers and block builders is an off-chain market enforced by software conventions: MEV-Boost relays sit outside the protocol, and the chain itself has no knowledge of the deals made there. Under Glamsterdam, the protocol handles the sale of block space. A proposer envisions a block, an enshrined beacon attests that payment will arrive, and only then is the block added to the chain. The team behind the full ePBS design has described the result as a neutral market where the only price signal is a market price, with no mandated relays, no off-chain payments, and no trust in side deals between builders and validators.
It is also a direct answer to a problem that has shown up in enforcement: a proposer who sells block space today cannot prove that they accepted the best offer, and mid-slot reorgs built around sold blocks are a known attack surface. Enshrining the mechanism gives the chain a way to check the deal directly, with the in-protocol beacon holding the payment until the block body is revealed to the network.
Block-level access lists, the second pillar, list in advance every account and storage slot a block touches. Nodes can execute blocks in parallel instead of sequentially, wallets get more accurate transaction preconfirmations by letting nodes predict the outcome of a transaction before it runs, and light clients can verify state access without processing every block. The upgrade also routes blob fees through a blob-based fee market, with fees for blob transactions burned rather than paid to proposers.
Rehearsals went through, client updates are due
September hardhat ramps on devnet already processed房子的 new block-building rules while testing limits up to 200 million, and the sequencing worked at each stage. The remaining work is operator-facing. Node operators running Sepolia must upgrade both execution and consensus clients before October 6, and the Foundation has said releases will be confirmed in a client release table.
The Foundation’s guidance to non-staking Sepolia operators: before October 6, 2026, update both your execution and consensus layer clients to releases that support the scheduled Sepolia activation.
Application developers face a longer checklist, because contract code that reads gas inside the EVM has always been sensitive to scheduling. Test suites should be re-run against the new rules, and gas estimation logic in wallets and infrastructure tools needs review before mainnet. Contracts that rely on fixed gas stipends, hardcoded limits, or assumptions about remaining gas may need changes. Application developers can test application behaviour in advance, either by using a Sepolia hardhat ramp in a local environment or by pointing at one of the public devnets where the upgrade is already live.
The wider rollout assigns each part of the fork a role. Fusaka counted blob capacity increases through the BPO schedule, and Glamsterdam picks up the same thread by raising per-block execution capacity on both execution and consensus layers. Gloas adds a path to a certified preconfs standard, which opens application design where early assurances are priced separately from settlement. Application developers and node operators can also test application behaviour in advance against the new rules on Sepolia before mainnet arrives.
What mainnet timing depends on
Sepolia’s gas ceiling climbs only if node operators and staking services choose it. Proposers who prefer not to run blocks near 200 million can keep the lower default, and that split will show up in real block data rather than in a roadmap. Devnets have carried the load so far, and none of them carries economic value or enough user traffic to stress relay timing, block propagation and builder competition at the sizes promised for mainnet. Sepolia is the first public network to try the configuration with real operators, real builders and an audience watching.
The mainnet window has not been fixed, and the Foundation has said a date will be announced as the fork approaches. The risk side is equally plain: larger blocks mean heavier network load, and block propagation at 200M gas is exactly what the testnet is meant to measure. A duplicate mainnet proposal across slots is the known failure mode under current MEV-Boost tuning, which is why a misinclusion penalty for duplicate mainnet proposals is part of the path to mainnet. Developers are also tracking the throughput export voucher, which mainnet adoption will hinge on, and state growth, since state creation gets more expensive as the network scales to the higher floor.
In this sense Sepolia becomes the last stop before a mainnet launch that developers have watched move through devnets for months. A clean fork on Tuesday keeps the Q4 mainnet window alive. A messy one pushes it.