Glamsterdam on Sepolia: Clients Ready, 3 Missing

Glamsterdam on Sepolia client releases are out — one day before the scheduled date. All five execution clients have the Amsterdam fork set for Unix time 1791294816, that is Tuesday, October 6, 2026, 13:53:36 UTC. On the consensus side, Teku, Prysm, Lodestar and Erigon’s Caplin have caught up. For Lighthouse, Nimbus and Grandine, as of September 28, 19:00 UTC, there was still no stable release that schedules the Sepolia fork. Anyone running a Sepolia node with one of those three still has no version that will carry them past October 6.

Glamsterdam on Sepolia: release status on September 28

Checked against each client’s release pages. The dates below are when the release was published on GitHub.

Execution layer — complete:

Geth v1.17.6 (Sep 23) — “The Amsterdam hardfork is scheduled on the Sepolia testnet at timestamp 1791294816“
Nethermind 2.0.0 (Sep 22) — its own section “Ready to the Glamsterdam fork on Sepolia”, including fork block 11296768
Besu 26.9.0 (Sep 25) — fork scheduled, plus a Layered-Txpool tuning for the upcoming 200M gas limit
Erigon v3.7.0 (Sep 25) — Amsterdam in the execution layer and Gloas in Caplin, epoch 353024
Reth v2.7.0 (Sep 28) — “Sepolia operators must upgrade before 6 October 2026 at 13:53:36 UTC”, plus persistent RocksDB tables for block-level access lists

Consensus layer — four of seven:

Teku 26.9.1 (Sep 28) — declared a “required update for the Sepolia network”
Prysm v7.2.0 (Sep 28) — fork scheduled, with a gas-limit caveat (see below)
Lodestar v1.49.0 (Sep 28) — fork plus the 200M gas limit scheduled
Caplin, the consensus client inside Erigon v3.7.0
Lighthouse — latest stable release v8.2.2 from August 18, with no Glamsterdam reference
Nimbus — latest stable release v26.8.0 from August 28; nightlies still appear daily
Grandine — latest release 2.0.6 from August 3

What the configuration says

The values live in the eth-clients/sepolia repository — the configuration the clients actually load:

Amsterdam Time:     1791294816   # Tue, October 6, 2026, 13:53:36 UTC
Gloas Epoch:        353024
Gloas Fork Version: 0x90000076

GAS_LIMIT_SCHEDULE:
  - EPOCH: 353024
    GAS_LIMIT: 200000000

The two names in the config are not a typo. Like its predecessors, Glamsterdam is a portmanteau of the two layer fork names: Gloas on the consensus layer, Amsterdam on the execution layer. The EL fork activates on a Unix timestamp, the CL fork on an epoch — which is why both figures sit side by side.

The gas limit: a tripwire for Prysm validators

On September 24 a GAS_LIMIT_SCHEDULE per EIP-8261 landed in the Sepolia config (eth-clients/sepolia#130): from epoch 353024 the network gas limit rises to 200 million. On Sepolia the number is therefore no longer just an upper bound the protocol allows, but a scheduled value.

That entry arrived after Prysm’s release cut. The Prysm release notes are unusually detailed on this point, for good reason: v7.2.0 does not include the schedule, so validators on that version fall back to 60 million in their Gloas proposer preferences. Anyone who wants to propose with 200 million has to set the value explicitly — "gas_limit": "200000000" in a version-2 proposer settings file, or via POST /eth/v1/validator/{pubkey}/gas_limit on the keymanager API. One more detail: --suggested-gas-limit only applies to pre-Gloas mev-boost registrations and has no effect after the fork. A follow-up release is meant to ship the schedule.

Lodestar already has the schedule on board in v1.49.0.

The two headliners

Meta-EIP EIP-7773 lists 18 core EIPs as scheduled for inclusion. Two of them are the real structural changes.

EIP-7732: Enshrined Proposer-Builder Separation

EIP-7732 pulls the separation of proposer and builder into the protocol. Instead of a full ExecutionPayload, the beacon block will carry a signed SignedExecutionPayloadBid — a commitment from the builder. Payload validation moves to the next slot, so a slot can end up “full”, “skipped” or “empty”.

New pieces include a builder registry with a staking requirement (at least 1 ETH, tracked in the beacon state) and a Payload Timeliness Committee (PTC): a group of validators that attest whether the builder delivered on time, without fully validating the payload themselves. That turns attesting into work that no longer has to wait behind execution validation.

How much machinery hangs off this shows up in the release notes: Prysm ships a new builder-registry endpoint for Gloas (POST /eth/v1/beacon/states/{state_id}/builders), version-2 proposer settings, per-key builder configuration through the keymanager API, remote signing for Gloas blocks, and a Builder Circuit Breaker that bans builders who win an auction and then fail to reveal the payload.

EIP-7928: Block-Level Access Lists

EIP-7928 turns what EIP-2930 offered optionally into something mandatory and block-wide: a block declares up front which accounts and storage slots it will touch — including the values after execution.

In concrete terms:

a new header field block_access_list_hash with the Keccak-256 hash of the RLP-encoded list
pure reads are recorded as well, along with balance, nonce and code changes and their post-transaction values
a cap of block_gas_limit // 2000 entries
deterministic ordering: addresses lexicographically, changes by block index and access order
an Engine API extension with ExecutionPayloadV4

The point is parallel execution: parallel disk reads, parallel transaction validation, parallel state-root calculation — and state reconstruction during sync without replaying transactions. That this is no longer a paper EIP shows up in Reth: v2.7.0 enables persistent RocksDB storage for block-level access lists and adds flags to keep computed BALs in cache for repeated traces (--rpc-cache.cache-computed-bals, --rpc-cache.prewarm-bals).

Who needs to change what

Sepolia node operators. The most concrete point: upgrade before October 6, 13:53:36 UTC. Anyone who does not will stop following the chain from the fork timestamp. For execution clients the versions are already all there. On the consensus side, Teku, Prysm, Lodestar or Caplin are the options that work today — for Lighthouse, Nimbus and Grandine it is wait or switch.

Sepolia validators. On top of the upgrade, check the gas-limit setting as above. Prysm users who do not set the value explicitly will propose with 60 million after the fork instead of 200.

Infrastructure and tooling teams. ExecutionPayloadV4 and the new header field are breaking changes for anything that parses Engine API payloads or block headers itself — indexers, explorers, custom sync logic. Sepolia is the first public chance to test that against a real chain.

Staking and validator software. ePBS adds an extra validator duty with the PTC and changes withdrawal handling toward an asynchronous flow between consensus and execution. That is more than a parameter update.

Smart-contract developers. Beyond the headliners, Glamsterdam includes several repricing EIPs, among them EIP-8037 (State Creation Gas Cost Increase) and EIP-8038 (State-access gas cost update). The Ethereum Foundation published a dedicated post on testing the repricing impact in August. We have also covered the later Hoodi hold caused by those gas changes in our piece on EIP-8037 and the broken gas assumptions.

The roadmap after that

From testing call ACDT #97 on September 21:

September 29, 2026 — target date for client releases; most shipped early
October 6, 2026, 13:53:36 UTC — fork on Sepolia, epoch 353024
October 8, 2026 — go/no-go for Hoodi; ACDE #247 is also scheduled that day
October 27, 2026 — Hoodi, provisional

There is still no date for mainnet. The roadmap page on ethereum.org lists Glamsterdam as “Q4 2026 · Date not yet confirmed”.

What remains open

Three things belong in an honest assessment.

First: The Sepolia date has already slipped several times. A written timestamp is an intention, not a fait accompli — it can slip again before October 6. That the releases are now out makes it more likely, but not certain.

Second: The three missing consensus clients are the real open item. As long as Lighthouse, Nimbus and Grandine have no release with the Sepolia fork, part of the network lacks the foundation — and Sepolia is a proof-of-authority network with a limited validator set that is meant to surface exactly such gaps. What that means for the go/no-go on October 8 will be decided on the call, not here.

Third: Both headliner EIPs are formally not final. EIP-7732, EIP-7928 and meta-EIP 7773 are all in Review. That is normal for this phase, but it means detail changes are still possible.

Sources

As of September 28, 2026, 19:00 UTC. Release status checked against the clients’ release pages. This article reports on a protocol change and is not investment advice. English version published November 9, 2026.