Since we first announced Binius a little over a year ago, we have been hard at work refining and enhancing the open-source, high-performance binary proof system. Today we are excited to announce the alpha release of the Binius library and unveil the new binius.xyz documentation website. Developers are now free to experiment with and build on Binius (the website’s “Building” section details clearly how to get started). At the same time, we are launching the very first Binius application: an Ethereum state proving service. Built with the Binius Rust library and Reth Ethereum node, the live service generates compressed proofs of account inclusion in Ethereum’s Merkle-Patricia state trie (MPT) and can be downloaded for verification here.

Ethereum Stateless Clients

This proof-of-concept service demonstrates a practical stateless client solution for Ethereum. Full stateless clients are crucial for achieving true protocol decentralization and were highlighted by Vitalik Buterin as a central part of the roadmap in a series of recently published blog posts. Stateless clients allow nodes to verify blocks without persisting the entire Ethereum state, which imposes an expensive storage burden on full-node operators. The stateless model therefore lowers barriers to participation in Ethereum validation, allowing for nodes to run unnoticeably on consumer laptops, mobile devices, and eventually smartwatches.

Ethereum’s Merkle-Patricia trie storage mechanism has historically been extremely unfriendly to cryptographic proving. This is why Ethereum’s roadmap for implementing stateless verification factored in a shift to alternative data structures, either Verkle trees or Poseidon Merkle trees. However, Verkle trees lack quantum resistance and SNARK/STARK-optimized hash functions like Poseidon are less battle-tested than the robust SHA-3 standard functions.

Our live service showcases that Binius can do what no other proof system can today: enable a stateless client solution for Ethereum without necessitating cardinal changes to the protocol. Only a few minor improvements are required. Ultimately, Binius gives Ethereum a faster yet more conservative pathway to achieving its Verge development plan.

Our Binius-Based Solution

A stateless client requires each block to be accompanied by a list of account addresses and corresponding values (e.g., balance) read or updated during its execution and a succinct cryptographic proof that the values are accurate. Irreducible’s MPT proof compression service addresses the most challenging aspects of generating this proof. When each block is published, we retrieve the MPT data for all relevant accounts and use Binius to prove their validity. Produced with a mean latency of ~11 seconds, a proof for ~1,400 account reads compresses ~3 MiB of MPT data to ~1 MiB, reducing the bandwidth requirements of a stateless client solution by ~3x. While the current implementation falls short of a complete stateless client solution, which would need to prove the MPT update rules in addition to read rules and storage trie proofs in addition to account trie proofs, it demonstrates the viability of our approach.

Our technology expresses the account validation rules in M3, a powerful new arithmetization model for programming SNARKs. The M3 logic efficiently captures the complex Merkle-Patria trie operations without incurring the overhead of a general-purpose zkVM. For comparison, the highly optimized and open-source SP1 proof system can generate similar proofs by emulating 100 million RISC-V cycles, using its custom Keccak-256 precompile and compression feature. Generating the SP1 proof on a GPU-equipped AWS server takes ~5 minutes and results in a 1.2 MiB proof*.

| Prover  | Hardware                  | Accounts   | Proof Size (MiB)  | Proving Time (s)  | Verification Time (s)  |
| :------ | :------------------------ | :--------- | :---------------- | :---------------- | :--------------------- |
| SP1     | AWS g6.2xlarge GPU        | 1,419      | 1.2               | 309               | 0.088                  |
| Binius  | AWS c7a.4xlarge CPU       | 1,419      | 1.0               | 28.0              | 0.734                  |
| Binius  | AWS c7i.16xlarge CPU      | 1,419      | 1.0               | 12.83             | 0.679                  |
| Binius  | Irreducible AMD 9354 CPU  | 1,419      | 1.0               | 11.02             | 0.678

Note: SP1 is benchmarked on an AWS GPU instance (g6.2xlarge) costing ~$1/hr. Binius is benchmarked on two different AWS instances; the c7a.4xlarge instance price matches at ~$1/hr and the c7i.16xlarge instance matches the 64 vCPU core count of Irreducible’s AMD 9354 CPU.

We benchmark our service performance using a large block (1,419 accounts) representative of the size that a full stateless client would need to process. Here we provide a Grafana snapshot of faster proof times and smaller proof sizes for smaller, average-case blocks.

It's Time to Build

This is just the first step for Binius. With the feature set of our latest Binius release, we are excited to work with ZK teams to bring the benefits of high-performance proving to their protocols. We invite everyone to learn more about how Binius works at binius.xyz. As we continue development, we are also working on end-to-end FPGA-based hardware acceleration that will provide real-time proving solutions for Ethereum’s SNARKification roadmap.

* A note on SP1 benchmarks: we included the benchmarks against SP1 as a comparison because not many ZK projects are developer-friendly enough to solve the MPT verification problem, and SP1 is a high quality project that is capable of doing so. The performance gap between Binius and SP1 takes into account the fact that our solution directly addresses MPT verification without the overhead of a general-purpose VM. For transparency, you can find our code for SP1 in our sp1-mpt project on GitLab.