Stwo

  • Name: Stwo
  • URL: https://github.com/starkware-libs/stwo
  • Category: proving-system infrastructure / Circle-STARK implementation / Cairo-adjacent proof backend
  • Summary: Stwo is a lower-layer proving framework, not just the Starknet prover. The useful mechanism is a flexible AIR-style frontend paired with a Circle-STARK backend over the Mersenne31 field, plus multiple entry points from simple Cairo proof generation to custom AIR work. That makes it worth tracking as a mechanism note about backend choice, field assumptions, and how much of the stack developers are actually allowed to touch.
  • What it does:
    • Provides a Rust implementation of a CSTARK prover and verifier built around Circle-STARK ideas and Mersenne31-based performance goals
    • Exposes a frontend for expressing custom constraints and a backend for generating fast STARK proofs, rather than limiting usage to one fixed VM story
    • Integrates with Cairo so developers can prove Cairo execution through higher-level tools while still exposing a lower-level proving library and core framework
    • Offers multiple integration layers, including one-command proving via Scarb, Cairo-oriented proving libraries, and a core library for custom AIRs and research-grade proving systems
    • Publishes benchmarks, open-source code, and formal research material, including a dedicated Stwo whitepaper and the underlying Circle STARK research context
  • Key claims:
    • The repo README describes Stwo as a next-generation implementation of a CSTARK prover and verifier written in Rust and highlights Circle STARKs, high performance, and flexibility as core features.
    • The S-two Book introduction gives the clearest concise architecture summary: a flexible constraint frontend, a backend leveraging Circle STARKs over the Mersenne31 prime field, and seamless Cairo integration.
    • StarkWare’s 2.0.0 launch post shows that Stwo now spans several developer layers, from scarb prove for simple Cairo execution proofs to a Cairo library and a core library for custom AIRs. That decomposition is more analytically useful than filing Stwo as a generic prover brand.
    • The 2026 whitepaper formalizes Stwo as a Circle-STARK system over Mersenne31, introduces the flat AIR model, and highlights cross-domain correlated agreement as an important soundness notion for multi-table proofs. This makes Stwo useful not only as an implementation entry but also as a distinct arithmetization-and-soundness comparison point.
    • The older Circle STARKs paper remains relevant because it explains the underlying circle-curve construction and the efficiency argument around M31 that Stwo inherits.
    • The project should still be handled carefully rather than treated as fully settled infrastructure. The public README explicitly says Stwo is a work in progress and is not yet recommended for production use.
  • Whitepaper: Stwo has a formal whitepaper (https://eprint.iacr.org/2026/532), but the most useful current primary-source packet for corpus work is the whitepaper together with the repo README, S-two Book introduction, developer launch post, and Circle STARK background paper; see ../whitepapers/stwo-primary-sources-2026-05-12.md.
  • Sources:

Internal linkages

  • Keep this note on the strongest nearby contrasts: plonky3, binius, and sp1.

  • Last reviewed: 2026-06-01 UTC