Sigma Prime’s Lighthouse v8.3.0-rc.0 is a Sepolia client for the Gloas consensus fork, not a mainnet Ethereum upgrade. The release, published 2026-10-01 04:47 UTC, tells mainnet operators to stay on v8.2.3 until a proper v8.3.0 ships. Sepolia operators must upgrade the beacon node and validator client, plus a compatible execution client, before 2026-10-06 13:53:36 UTC.
That is a testnet clock. Hoodi and mainnet still have no activation date in the Ethereum Foundation’s Glamsterdam schedule. I care about shipped mechanics, not a roadmap slide that treats Sepolia as “Ethereum is upgrading next week.”
CoinGecko via our site snapshot at had ether (ETH) at $2,667.21 (−1.34% over 24 hours) and bitcoin (BTC) at $84,492 (−0.35%). The tape is mid-$80k chop. It does not move this client deadline.
What happened
Lighthouse v8.3.0-rc.0 is a release candidate with Sepolia support for Gloas, the consensus-layer half of Glamsterdam (Amsterdam on the execution layer). Sigma Prime’s table lists Sepolia at fork epoch 353024, fork slot 11296768, activation 6 October 2026, 13:53:36 UTC. The notes say to update both the beacon node and the validator client, together with a compatible execution client. The Gloas fork is not yet scheduled on Hoodi or mainnet.
The same notes are blunt about mainnet: users should stick to v8.2.3 until v8.3.0 proper is released. v8.2.3 itself, published 2026-10-01 02:15 UTC, is a medium-priority mainnet security release. Sigma Prime is not disclosing the security content yet, to give operators time to upgrade. Staking users: beacon node medium, validator client low. That is a different job from the Sepolia candidate.
Nethermind 2.1.0, published 2026-10-01 22:43 UTC, tells operators to upgrade every Sepolia node to 2.1.0 before the same Glamsterdam instant: Tue, 06 Oct 2026, 13:53:36 UTC. Most nodes are described as an in-place upgrade with no resync, with breaking-change caveats for the flat backend with custom sync, poacore, Flashbots relays, Parity-style traces, and plugins. Highlights include final EIP-8037 state-gas rules, a TX_MAX_TOTAL_GAS_LIMIT transaction cap, and Engine API alignment for Amsterdam.
The Ethereum Foundation’s Glamsterdam testnet announcement matches the clock: Sepolia at epoch 353,024, slot 11,296,768, 2026-10-06 13:53:36 UTC (UNIX 1791294816). Hoodi and mainnet rows are TBD. The post says this announcement covers Sepolia and does not schedule a mainnet upgrade. The consensus-layer table lists Lighthouse 8.3.0-rc.0 among Sepolia-compatible releases. Node operators must update both execution and consensus clients before activation.
Gloas on Lighthouse also ships builder workflow for enshrined proposer-builder separation (ePBS). External builders live in builder_definitions.yml in the validator directory. On Sepolia, the scheduled default gas limit becomes 200 million gas at Gloas, unless a per-validator setting or an explicit --gas-limit wins. The Fast Confirmation Rule (FCR) can confirm blocks in 1–2 slots and feed that hash as the execution layer’s safe_block_hash, but FCR is disabled by default. Enable it only with --enable-fast-confirmation. Finalization does not change.
The beacon node automatically upgrades its database (including to schema v31 on startup). Downgrading to v8.2.x means stopping the node and using the v8.3.0-rc.0 binary to migrate back to schema v29 before Gloas. Downgrading after Gloas is not supported. Partial data columns are now enabled by default on all networks, including mainnet. Opt out with --enable-partial-columns false.
Context
Glamsterdam is Amsterdam (execution) plus Gloas (consensus). The Foundation’s headliners are ePBS (EIP-7732) and block-level access lists (EIP-7928), plus gas repricing that is supposed to track execution and state growth more honestly (EIP-8037 and EIP-8038 among others). That is the protocol story. The operator story this week is narrower: Sepolia is the first public network with a pinned timestamp, and the client tags that claim that timestamp are now on GitHub.
We already treated Glamsterdam as a break-test, not a vibe, when Platåberget was the named lab. Sepolia is the next public fork with a UTC second attached. It is still a testnet. Builders who skip it will meet ePBS duties, payload-timeliness votes, and 200 million gas assumptions for the first time on a later network, which is a sloppy way to learn.
Fee context, not a lead: DefiLlama’s Ethereum fees overview, pulled 2026-10-03 00:16 UTC from api.llama.fi/overview/fees/ethereum, showed about $15.40 million in 24-hour fees, up 24.13% versus the prior day, and about $86.49 million over seven days. Chain TVL on the DefiLlama chains endpoint at the same pull was about $53.32 billion for Ethereum. Quiet fee days do not cancel a fork clock. Busy fee days do not print a mainnet date that the Foundation has not set.
The last high-priority mainnet Lighthouse patch we covered was v8.2.2. v8.2.3 is the current mainnet security tag. v8.3.0-rc.0 is not a substitute. For the chain’s longer arc, see Ethereum history. For how we timestamp tape versus protocol events, use how to read a crypto market snapshot.
Rollups still settle to L1. A Sepolia Gloas miss does not halt layer-2 withdrawals on mainnet. It can make the ePBS and BAL rehearsal fail in public, which is the point of the testnet. Keep client notes in Ethereum news beside the older Platåberget Glamsterdam break-test, not in a price thread.
Our read
Stance: treat v8.3.0-rc.0 as a Sepolia Gloas binary with a hard UTC second, and treat v8.2.3 as the mainnet security line. Do not collapse those into one “Ethereum upgrade this week” headline. Show me the fee and the fork epoch, not a roadmap that borrows Sepolia’s date for mainnet.
The pairing with Nethermind 2.1.0 is the tell. Consensus and execution both named the same instant. Operators who bump only Lighthouse, or only Nethermind, are not “mostly ready.” They are half-forked. ePBS also means builder config is now a Sepolia homework assignment, not a blog metaphor. If you propose with the wrong gas limit after Gloas, that is a settings miss, not a mystery.
FCR is interesting and off by default. Anyone marketing “1–2 slot safe” as the new Ethereum confirmation rule on mainnet is ahead of the release notes. Partial data columns defaulting on, including mainnet, is a real bandwidth change on the current tag family, but it is still not Gloas on mainnet.
Falsifiable claim: by 2026-10-06 23:59 UTC, either Sepolia Gloas/Glamsterdam activates near the published 2026-10-06 13:53:36 UTC instant on the Lighthouse v8.3.0-rc.0 plus Nethermind 2.1.0 pairing (or another Sepolia-tagged pair the Foundation lists), or Sigma Prime / the Ethereum Foundation publish a delay, cancellation, or retarget of that Sepolia time. Separately, if Sigma Prime’s v8.3.0-rc.0 notes are rewritten so that mainnet operators are told to run the rc instead of staying on v8.2.3, the “not a mainnet upgrade” half of this read is wrong.
What to watch next
Watch the Sepolia head around 13:53 UTC on 6 October 2026. A clean epoch 353024 transition on upgraded pairs is the pass. A stalled epoch, missing payload reveals, or a client advisory in the hours after is the fail. Screenshot the client versions you actually ran, not the blog table.
Watch Sigma Prime’s mainnet line. If v8.3.0 ships without “rc” and still says mainnet is not Gloas, the split holds. If a security advisory lands on v8.2.3, that is a mainnet patch story, not a fork-date story.
Watch builder and gas-limit settings on Sepolia after Gloas. Lighthouse documents a 200 million scheduled default. Other clients in the Foundation table warn that some default to 60 million unless you set proposer gas explicitly. Mix those defaults and you will misread “the network gas limit.”
Ignore ether’s 24-hour percent as a Glamsterdam signal. The snapshot at 2026-10-03 00:04 UTC is context. The fork is a client clock.