Arithmetic-friendly hashes have long been one of the weakest pieces of prover cryptography. With ZisK v1.3.0, we removed Poseidon entirely and moved to BLAKE3. What surprised us is that we got the same performance — or even better. After many optimizations elsewhere in the prover, hashing had become one of the most time-consuming parts of proof generation. BLAKE3 is more than an order of magnitude faster, making this cost almost disappear and largely compensating for the increased complexity of the recursion circuit. Getting there required a lot of re-engineering: circuit sizes, prover scheduling, and architecture all had to be reconsidered around the new trade-offs that BLAKE3 gives us. The final result surprised even us. There is a lot of focus today on binary-field proving. Binary fields attack the hash-proving bottleneck from a different direction, but introduce different trade-offs when proving regular arithmetic. There is a huge amount of research happening here, and at ZisK we are following it very closely. I'm genuinely curious to see whether binary-field architectures will outperform what we can achieve with the current approach — and how long it will take. So I like to think of v1.3.0 as setting a new target. 🎯 To everyone working on binary fields — and to all of us trying to make provers faster: let's see if we can beat it :) And we still have a few optimizations in the basket.