Glamsterdam Explained: Every EIP in Ethereum's Next Upgrade

Glamsterdam is the next Ethereum hard fork after Fusaka, and it is shaping up as the most structural release the network has shipped in years. It moves the block-building auction into the protocol itself, puts a consensus-verified state diff inside every block, and reprices the corners of the EVM that stood in the way of much higher gas limits. Multi-client devnets have been running the full slate since June 2026, the public Platåberget testnet forked on August 20, Sepolia forks on October 6, and ethereum.org puts mainnet activation in Q4 2026.

This page is the hub for the whole fork: what Glamsterdam is, where every EIP stands, and where to go deeper on the changes that matter most. Fork scope only becomes truly final when client teams cut mainnet releases, so we keep this page updated as the schedule firms up. Everything below is checked against the fork’s Meta EIP, EIP-7773, as of October 1, 2026.

What Glamsterdam is and where it sits

The name follows Ethereum’s fork-naming tradition. Consensus-layer upgrades are named after stars and execution-layer upgrades after Devcon cities, and since the two layers began forking together the names have merged into portmanteaus: Shanghai plus Capella became Shapella, Prague plus Electra became Pectra, Fulu plus Osaka became Fusaka, and Gloas plus Amsterdam becomes Glamsterdam.

Fusaka, which activated on mainnet on December 3, 2025, was a data-availability fork. Its headliner, PeerDAS, let validators sample blob data instead of downloading all of it, which opened the door to much higher blob throughput for rollups. Glamsterdam turns the attention back to layer 1 itself: how blocks get built, how fast the EVM can execute them, and how much gas a block can safely carry. The ethereum.org roadmap page files it under scaling the L1, and the devnets are testing gas limit configurations reaching toward 200 million, more than triple today’s mainnet limit.

Glamsterdam is also one station on a much longer line. The protocol’s multi-year direction, with lean consensus, post-quantum signatures, and a slimmer state model, is a story of its own; we cover it in our lean Ethereum roadmap explainer. This post stays on the fork that is actually next.

The timeline

Devnets ran the complete Glamsterdam feature set from June 2026, and the fork has been live in public since August 20, when Platåberget, a community-joinable EthPandaOps testnet, forked at epoch 1536. Platåberget is built to run for a few months rather than days, so there is a stable place to test post-Glamsterdam behavior right now, including public validator and builder deposits for anyone who wants to rehearse the ePBS builder workflow. The next step is the long-lived testnets, Sepolia and Hoodi, where slot timings, gas schedules, and the remaining parameters get their last tuning under realistic conditions. Mainnet comes after those forks hold stable.

Sepolia now has a firm date. The Ethereum Foundation’s testnet announcement schedules it for epoch 353024, slot 11,296,768, on October 6, 2026 at 13:53:36 UTC. The epoch sits in Sepolia’s client configuration and in the Meta EIP’s activation table, and client teams have shipped Sepolia-compatible releases, among them Geth 1.17.7, Nethermind 2.0.0, Reth 2.7.0, Besu 26.9.0, Prysm 7.2.0 and Teku 26.9.1, with Lighthouse and Grandine on release candidates. Hoodi is next but not scheduled: developers have pencilled in October 27 and will make the call after watching Sepolia fork. Mainnet has no date. ethereum.org lists activation for Q4 2026, and the EF’s announcement says plainly that it does not schedule a mainnet upgrade; that comes in a separate announcement once Sepolia and Hoodi hold stable. You can watch the clock tick toward Sepolia on our live Glamsterdam countdown, which we update as each epoch lands.

The scope at a glance

Ethereum forks are assembled in public, and the vocabulary matters for reading any Glamsterdam news. An EIP is first Proposed for Inclusion, becomes Considered for Inclusion (CFI) once client teams agree it is a serious candidate, and is Scheduled for Inclusion (SFI) when it is committed to the fork and implemented on devnets. Scope firmed up over the summer: eighteen EIPs are now scheduled for Glamsterdam, up from the ten committed when the full-slate devnets started:

The shape of the list tells its own story: one enormous consensus change, one enormous execution change, and sixteen smaller EIPs that mostly exist to make the fork’s throughput ambitions safe, with a few, like cheaper plain transfers and faster validator exits, that are simply good on their own.

EIP-7732: the block auction moves into the protocol

Today roughly nine out of ten validators outsource block building through MEV-Boost, a system held together by unpaid, trusted relays that shuttle blind headers between builders and proposers and vouch for payment. EIP-7732, enshrined proposer-builder separation, moves that entire market on-chain. Builders become staked entities in their own beacon chain registry, their signed bids commit to an exact payload hash, and payment is deducted from their on-chain balance the moment a proposer includes the bid, whether or not the payload ever appears. With payment enforced by the protocol, nobody needs a relay to escrow trust.

The consequence that ripples furthest is timing. The beacon block stops carrying the execution payload; the payload is revealed mid-slot and attested by a new Payload Timeliness Committee of 512 validators, while regular attesters vote on the consensus block without executing anything. That stretches the propagation-critical window from about two seconds to about nine, which is what allows blocks to get much bigger, and it compresses the window in which builders assemble blocks on fresh state to roughly four to six seconds. For builders it means posting real capital and learning a new auction; for searchers it means every deadline in the pipeline moves earlier. We take the whole change apart, slot anatomy included, in our ePBS deep dive.

EIP-7928: every block ships its own state diff

The execution headliner is quieter but arguably more practical. Under EIP-7928, every block carries a Block-Level Access List: a complete, consensus-verified record of every address the block touched, every storage slot written with its post-value, every slot read, and every balance, nonce, and code change, each keyed to the transaction that caused it. The block header commits to the list’s hash, and every validating node rebuilds it during execution; a block whose list does not match reality is invalid.

The protocol wants BALs because they remove the information barrier that forces the EVM to run transactions one at a time. With access sets declared and enforced up front, clients can prefetch state, execute non-conflicting transactions in parallel, and split state-root computation across cores, which is the groundwork the 200 million gas ambition rests on. Data consumers should want them for a different reason: a verified state diff in every block replaces a whole category of expensive trace-based re-execution. The change is further along than most people realize; Geth already builds and verifies BALs ahead of the fork. What is in a BAL, what it replaces, and what it deliberately does not do is covered in our Block-Level Access Lists explainer.

The repricing slate that makes 200M gas thinkable

Five of the scheduled EIPs are gas repricings, and they read as a single project: close the gaps that would make a 150 to 200 million gas block dangerous before raising the limit toward those numbers.

EIP-7778 changes how gas counts toward the block limit. Refunds, like the ones earned for clearing storage, still reduce what a transaction pays, but they no longer shrink the transaction’s footprint in the block’s accounting, so a block can no longer pack more real execution work than its limit implies. EIP-7976 raises the floor cost of calldata, continuing the work EIP-7623 started in Pectra, which keeps the worst-case size of a data-stuffed block bounded as the limit climbs. EIP-7981 raises the price of EIP-2930 access lists so that declaring state access costs something closer to what it makes every node actually do. And EIP-8037 makes creating new state, fresh accounts and fresh storage, more expensive, because state growth that is manageable at 60 million gas per block becomes a flood at three times that. It also splits gas into two dimensions: state creation is now metered at a fixed cost per state byte and charged at runtime from a separate state gas reservoir, which is why a plain ETH transfer is no longer always a flat 21,000 gas; sending to an account that does not exist yet incurs state-creation gas on top. EIP-8038, promoted into the scope over the summer, completes the slate by raising the price of state-access opcodes like EXTCODESIZE and EXTCODECOPY, which have been underpriced relative to the two database reads they force on every node.

None of these are exciting on their own, and that is the point. The gas limit itself is not set by the hard fork; validators vote it block by block. The repricing slate is what makes voting it sharply upward defensible.

Quality-of-life changes for contract developers

Another cluster of scheduled EIPs lands closer to everyday contract work. EIP-7954 lifts the maximum size of deployed contract code from 24 KiB, a cap that has sat unchanged since 2016 and forced large protocols into awkward proxy-and-library contortions, to 64 KiB, with maximum initcode growing from 48 KiB to 128 KiB. EIP-8024 adds SWAPN, DUPN, and EXCHANGE, stack instructions that reach deeper than the sixteen slots today’s SWAP and DUP can touch; they were designed for the EOF project, and this EIP salvages them in a backward-compatible form that compilers can use to bury “stack too deep” errors. EIP-7843 adds a SLOTNUM opcode that exposes the current beacon chain slot to the EVM, giving contracts a native clock at slot granularity instead of inferring it from timestamps. And EIP-2780, another summer promotion, cuts the intrinsic cost of a simple transfer between existing accounts by up to 71 percent by decomposing the flat 21,000 gas into the work a payment actually does, with a surcharge only when the recipient account has to be created.

EIP-7708 may be the sleeper of the fork for anyone running data pipelines: every ETH value transfer, including plain sends and transfers made inside contract calls, emits a log. Today native ETH movements are invisible to log-based indexing, and catching internal transfers means tracing entire blocks. After 7708, ETH becomes trackable the way ERC-20 tokens always have been, with a log per transfer, which collapses a lot of tracing infrastructure into a simple subscription; the pipeline pattern is the one described in our Ethereum webhook guide.

EIP-7997 rounds out the developer-facing list: a minimal deployment factory baked into the protocol at a fixed address, so wallets and applications can hold the same address on every EVM chain natively instead of relying on CREATE2 workarounds. EIP-8246 removes the SELFDESTRUCT burn, another step in retiring that opcode.

What left the bucket, and what got cut

The Considered for Inclusion bucket is empty. Of the nine maybes this page listed in July, eight were promoted into the scope (2780, 7688, 7997, 8038, 8045, 8061, 8246 and 8282), and the ninth, EIP-7610 (revert contract creation into non-empty storage), no longer appears in the Meta EIP. ethereum.org now describes the scope as frozen, with the caveat that the Meta EIP stays in Review until the fork ships, so a testnet failure could still force a removal. The consensus additions beyond ePBS are housekeeping with real consequences: EIP-8045 stops slashed validators from being picked to propose, eliminating guaranteed missed slots during mass-slashings, and EIP-8061 lets exit capacity scale with total stake, roughly quadrupling withdrawal throughput at current levels while halving the weak-subjectivity window to about seven days.

A fork’s outline is also drawn by its networking layer, which the headline table omits: eth/70 (EIP-7975, partial block receipt lists), eth/71 (EIP-8159, block access list exchange between peers, now required of all execution clients), eth/72 (EIP-8070, sparse blobpool), cell-level deltas for data column broadcast (EIP-8136), and BAL-based state healing (EIP-8189) all ship alongside, plus two informational EIPs, 7904 and 8261, that document the gas cost analysis and the gas limit schedule.

The two famous declines still explain the fork’s shape. FOCIL, EIP-7805, the fork-choice-enforced inclusion list design, lost the consensus headliner seat to ePBS in 2025, when the debate came down to censorship resistance first versus restructuring the block pipeline first. Declined here did not mean dead: the EF’s Hegotá tier list, published on September 7, names FOCIL the next fork’s must-ship consensus headliner; we cover that in our Hegotá explainer. EOF, EIP-7692, was declined here as in earlier forks, with EIP-8024 rescuing its most-wanted opcodes.

What Glamsterdam means for latency-sensitive users

If you run a searcher, a trading bot, or anything else that races the chain, the fork’s changes point in one direction: more data per block, less time per decision. ePBS compresses the window for building on fresh state to a few seconds and makes the mid-slot payload reveal the moment that matters, so every network round-trip in your loop gets more expensive relative to the budget. BALs add tens of kilobytes to every block, and a gas limit heading toward 200 million multiplies both block size and the value of reading it without crossing the internet. The fork rewards code that sits where the data lands. One warning from the EF’s Platåberget announcement deserves emphasis even if you never touch MEV: the repricing bundle breaks any tool that hardcodes a maximum block gas limit (wallets, indexers and gas estimators are named explicitly) and any code that assumes a plain transfer always costs 21,000 gas. Platåberget and Sepolia exist precisely so this class of breakage gets found before mainnet.

That is the model BLAZED.sh is built on: your containers and scripts deploy onto hosts running fully synced Ethereum nodes and talk to them over a local socket, with sub-10ms RPC round-trips and an unfiltered view of the mempool from the node your code sits on. Ethereum mainnet is the only chain live on the platform today, and it happens to be exactly the chain Glamsterdam rewires. The window before the fork is the right time to move latency-sensitive workloads next to the node, rehearse on Platåberget today, on Sepolia from October 6 and on Hoodi when it forks, and let everyone else discover the new timing constraints on mainnet.

FAQ

When does Glamsterdam go live?

Mainnet activation is targeted for Q4 2026, but no mainnet epoch is set. The fork is already live on Platåberget, the public testnet that forked on August 20, 2026. Sepolia follows at epoch 353024 on October 6, 2026 at 13:53:36 UTC, and Hoodi is tentatively planned for October 27, pending how Sepolia goes. Our Glamsterdam countdown tracks the next activation.

What are the headline changes in Glamsterdam?

EIP-7732, enshrined proposer-builder separation, on the consensus layer, and EIP-7928, Block-Level Access Lists, on the execution layer. The first moves the block-building auction into the protocol; the second puts a consensus-verified state diff in every block and lays the groundwork for parallel execution.

Is the Glamsterdam EIP list final?

Eighteen EIPs are Scheduled for Inclusion and running on a public testnet, the EF’s Sepolia announcement lists the same eighteen, and ethereum.org describes the scope as frozen. The nine-EIP maybe bucket that existed in July is empty: eight entries were promoted, one dropped out. Details can still move until Sepolia and Hoodi fork and settle, and this page is updated when they do.

Does Glamsterdam raise the gas limit?

Not directly; the gas limit is voted by validators, not set by forks. But the slate is built to make a much higher limit safe: BALs unlock parallel execution, ePBS extends the propagation window, and the repricing EIPs close the worst-case vectors. Devnets are testing configurations reaching toward 200 million gas. The vote is visible in the client releases: Prysm 7.2.0 and Teku 26.9.1 default validators to a 60 million limit after the Sepolia fork, and operators who want to propose 200 million blocks have to configure it explicitly.

Conclusion

Glamsterdam is two big ideas and a supporting cast: put the block auction in the protocol, put a state diff in every block, and reprice whatever stands between Ethereum and a much higher gas limit. Eighteen EIPs are locked, the maybe bucket is empty, and the calendar now reads Platåberget live, Sepolia on October 6, Hoodi tentatively on October 27, mainnet in Q4 2026. We will keep this page current as the remaining epochs land; for the deep ends of the two headliners, start with our ePBS deep dive and the Block-Level Access Lists explainer, and for what comes after, see our Hegotá tier-list explainer.