Stone and S-two -- what's the leap we're talking about? Stone is StarkWare's first production grade prover. Some numbers: - It has been running in production for over 5 years - It has settled over $1T - It saved over $1B in fees for users It's a beast. S-two is the second time we're building a prover. As you'd expect, second time you build something, there are a lot of improvements. The biggest improvement is the size of numbers used to construct the proof (field size). In Stone we used 256-bit numbers, and S-two uses 32bit numbers. Comparing them side by side, multiplication (a very important and costly operation in STARKs) is 120x faster! Let me repeat that: 120X faster. 🏎️ But there are a lot of other innovations under the hood. For instance, rather than working with one step at a time, and all steps are of the same nature, we now have nearly unlimited flexibility to add various opcodes and operations in a very efficient way that leads to further improved performance (the S-two whitepaper will give more details on this). Modularity is another cool aspect -- the actual proof generation can be interfaced in a modular way with various AIRs (constraint sytsems). This allows external teams like @NexusLabs , @KakarotZkEvm and @gizatechxyz to use S-two for computations that are different than what we use them for. So S-two ends up the fastest prover in the world, more than 100x faster than Stone. Will the next generation prover, coming out a few years down the line (no, we're not working on it yet) be again 100x faster? Honestly, I doubt that, but at same time I also didn't expect the 100x factor we got so far.