We are releasing OpenVM v1.4.1 today under MIT and Apache 2.0 dual-license on Github. This release introduces a new Virtual Pool Memory Manager in the GPU prover as well as several memory access optimizations in our Cuda kernels. These provide substantial performance gains, dropping average latency for Ethereum mainnet blocks under 12 seconds and enabling proving cost as low as $0.0001 per transaction.
To evaluate, we generated STARK proofs for Ethereum mainnet blocks using L40S GPUs on 64 g6e.2xlarge instances on AWS. We measured end-to-end proving times as well as proving cost per transaction. On blocks 21,000,000 to 21,000,003, OpenVM v1.4.1 shows marked improvements in latency and cost.
We also generated proofs for the 1000 block range of mainnet blocks between 22,000,000 and 22,000,999 to compare with OpenVM v1.4.0 from September 2025, which show these cost and latency improvements persist across a wider range, achieving average case real-time proving over this range.
OpenVM v1.4.1 has been audited by Cantina and is recommended for production usage. Day 1 hosted proving support is also available through Axiom on the Axiom Proving API. Check out our docs to get started with OpenVM, and stay updated via our developer chat on Telegram and the code on Github.