# Vitalik Buterin proposes EIP-8288 to make Ethereum cryptography cheaper

> Buterin backs EIP-8288 over new precompiles: aggregated STARK proofs off-chain and post-quantum signatures far cheaper.
> By José Maldonado · Ethereum · 2026-10-21

For years Ethereum has solved every new cryptographic need with a precompile: a function embedded in node software that makes a very specific operation cheaper in exchange for a protocol change. Vitalik Buterin wants to break that inertia. On October 19, 2026, [he defended](https://x.com/VitalikButerin/status/2112032009413439704) on X what he himself called "EIP-8288 maximalism", a proposal he co-signed with researcher Thomas Coratger that moves heavy cryptography off-chain. By his calculations, a private quantum-proof transaction would go from about 10 million gas to tens of thousands.

The draft was filed on June 3, 2026 and has not been approved. Its core idea is quickly stated: the transaction does not carry the signature or the proof, it carries a 96-byte declaration saying what must be verified, and mempool nodes bundle those declarations into a single recursive proof every second. The block loads only that final proof. What is avoided is precisely what does not fit today: post-quantum signatures of 2 to 3 kilobytes and proofs of more than 128 kilobytes traveling as data in every transaction.

The interest is not only in the gas bill. With EIP-8288, Ethereum would stop deciding which cryptographic schemes it supports: verification is encapsulated in a proof the protocol does not need to understand. The Foundation's official plan reserves post-quantum aggregation for fork M*, while Buterin wants it in I*.

## A precompile weighs forever

A precompile does not run like a normal contract: it lives inside every execution client, from Geth to Erigon, and makes cheap operations that in pure bytecode would cost too much. The problem is that adding one requires a hard fork and an identical implementation in every client. Ethereum saw this in September 2016, when a wave of denial-of-service attacks exploited the precompiles and forced two emergency forks to recalibrate their price, as Mark Rutherford recalls in Tech Times.

![Vitalik Buterin and his push to make Ethereum cryptography cheaper](https://content.observatorioblockchain.com/wp-content/uploads/2026/10/vitalik-ether-precompilado-crypto.webp)
*Vitalik Buterin and his push to make Ethereum cryptography cheaper*

Buterin had long been trying to close that door. In 2024 he proposed turning the identity precompile into virtual machine code, and a later draft did the same with five little-used precompiles.

EIP-8288 changes the unit of work. The user proves locally the validity of their signature inside a STARK proof and sends a 96-byte dependency with the scheme, the data hash and the key hash.

Mempool nodes fold those dependencies every second and the block includes a single recursive proof covering all transactions. In Buterin's words, precompiles are "the wrong place" to sustain Ethereum's cryptography, and with this model the network "would never have to add another precompile".

The move fits with the rebuilding of tools Ethereum has already been doing for its post-quantum stage, work this outlet has followed in explaining [how post-quantum cryptography pushes a new era of engineering in Ethereum](https://observatorioblockchain.com/analisis-sociologicos/ethereum-impulsa-la-nueva-era-de-la-ingenieria-postcuantica/). The difference is one of degree: the debate is no longer which algorithm to choose, but whether the protocol should keep choosing them.

## The real cost, from 10 million gas to tens of thousands

The draft's own figures set the savings. A hash-based post-quantum signature measures between 2 and 3 kilobytes and costs between 150,000 and 200,000 gas to verify. A standalone STARK proof exceeds 128 kilobytes and reaches 512 kilobytes when generation speed is prioritized, which takes it to millions of gas. The specification sets 3,000 gas per signature [leanSPHINCS](https://github.com/leanEthereum/leansphincs) declared and 30,000 for the recursive proof, so a transaction with a single dependency is around 33,000 gas.

Operation

Cost today

With EIP-8288

Hash-based post-quantum signature (2 to 3 kB)

150,000 to 200,000 gas

about 33,000 gas

Standalone STARK proof (128 to 512 kB)

millions of gas

included in the block proof

Private post-quantum transaction

close to 10 million gas

tens of thousands

Standard Ethereum transaction

about 21,000 gas

unchanged

The "more than 99%" repeated by coverage on October 19 and 20 rests on the extreme case, that of private post-quantum transactions, which today approach 10 million gas and would fall to tens of thousands according to the calculations Buterin published in September and Decrypt picked up. For a standalone signature, the cut implied by the draft's parameters is around 80%, not 99%. Tech Times attributed the percentage to signature verification in general.

The draft also limits the promise: each transaction allows 16 signature dependencies and one STARK proof, and repeating a dependency is paid again. On the builders' side, aggregation implies about 256 kilobytes of fixed proof every second.

## Who decides which cryptography the network supports

Here is the point that has been least discussed. The Ethereum Foundation's post-quantum roadmap, published at pq.ethereum.org, chains four milestones: key registration in I*, post-quantum signature precompiles in J*, attestations and consensus proofs in L*, and only in M* signature aggregation. The protocol cluster plan, dated September 7, 2026, promises a quantum-resistant L1 across all three layers by December 2029.

The timelines of that path are the ones [Thomas Coratger detailed,](https://observatorioblockchain.com/en/blockchain/ethereum-quantum-resistance-thomas-coratger/) with Lean Ethereum as the common substrate between consensus and execution, and with signature aggregation as the final stage.

Buterin signs on to that same technical direction and disagrees on the order: he wants EIP-8288 in I\*, before any post-quantum precompile is installed. If it succeeds, any team able to write a STARK circuit could deploy a new signature scheme without submitting an improvement proposal or waiting for a hard fork.

That shift of power barely surfaced in the public debate. Tech Times notes there is no formal position on whether distributing innovation outside the proposals bottleneck is an advantage or a risk of schemes arriving without review. The precedent of EIP-8361, when [Aave came out against the proposal and against the Foundation's ivory tower](https://observatorioblockchain.com/en/ethereum/aave-carga-contra-la-eip-8361-y-la-ethereum-foundation/), shows that this tug-of-war is not new.

## Hash versus lattices, the fight that pushed the proposal

The October 19 messages closed a weekend of discussion that began on Friday with the publication of Aerie, a system for aggregating lattice-based Falcon-512 signatures signed by LayerZero Labs with a16z crypto, Carnegie Mellon University and Imperial College London. In the demo it aggregated 32,768 signatures in 0.65 seconds on a laptop with an M4 Max chip and no GPU. Each aggregated signature takes up about 42.1 bytes.

a16z crypto's argument is that the recursive demonstration on which Ethereum's post-quantum plan relies is intricate and error-prone. Eli Ben-Sasson, CEO of StarkWare, responded that lattices are not mature enough to secure large assets, according to Bankless, and Tech Times notes that StarkWare has a direct commercial interest in STARK proofs being the standard verification layer.

EIP-8288 does not close the debate, it shelves it. Anyone using a scheme other than leanSPHINCS will bear the on-chain cost or wrap it in a STARK proof. Buterin acknowledges that, if lattices withstand AI-assisted attacks, precompiles for accounts and wallets could still come.

## Frames, I\* and a calendar without signatures

The draft is far from final. The aggregation route with leanSTARK appears as "reserved" in the specification, the parameters for the aggregated verification key and the dependency hash are still pending, and the reference implementation is listed as upcoming. It also depends on EIP-2718 and EIP-8141, known as Frames, which likewise has no fork assigned.

None of this activates by decree. Nodes without the upgrade would reject transactions with the new frame mode and blocks with the recursive proof header, so entry into mainnet requires a coordinated upgrade and the go-ahead from client teams on the development calls.

The proposal now enters that phase, the same one in which it will be decided whether J\* keeps its post-quantum precompiles or whether the official plan is rewritten to jump straight to signature aggregation.

What is at stake for Ethereum in this discussion is not a percentage of gas. It is whether the protocol's cryptography will keep growing on the basis of embedded functions and forks negotiated one by one, or whether it becomes software that anyone can write, prove and deploy on their own. The first option is slower and more predictable. The second leaves less control in the Foundation's hands and shifts the risk to circuits no one has audited yet.
