Releasing OpenVM 2.0 to Production

We are excited to release OpenVM 2.0 to production, combining state-of-the-art performance with enterprise-ready security and new features for proof aggregation. In conjunction with this production release, we are introducing a new security framework resting on three pillars:

• Human security review via an external audit by zkSecurity
• Expanded formal verification coverage for Keccak and SHA-2 in addition to RISC-V
• Continuous vulnerability scanning by a proprietary AI tool from Axiom

After completing this process, we feel confident recommending OpenVM 2.0 for production use.

New additions in this release include GPU proving for Halo2 wrapping and deferrals, a new feature that allows the verification of arbitrary ZK circuits in OpenVM. Deferrals allow proofs from custom circuits to be composed with general-purpose computation inside a single OpenVM proof. We are excited to see the new applications developers build with this feature.

On the performance front, OpenVM 2.0 further improves upon the Beta release in April. It now proves mainnet Ethereum blocks in real time on both 8 and 16 GPUs, with p99 times of 9.8s and 6.3s respectively, while maintaining 100 bits of provable security and sub-300kB proof size.

‍

OpenVM 2.0 is open source under the MIT and Apache 2.0 licenses on GitHub, including the GPU prover. To learn more, check out the developer docs or follow us on X.

Enterprise-Ready Security

We are introducing a new security framework as part of the production release for OpenVM 2.0 that combines three levels of assurance: traditional external audits, formal verification, and continuous vulnerability scanning by our proprietary AI tool. Specifically, OpenVM releases go through the following layers of security:

• Human security review: OpenVM 2.0 undergoes external security review as a standard process for production releases. It has completed a rigorous external audit by zkSecurity (report available here) and is recommended for production use.
• Formal verification: We have expanded the formal verification coverage of the system to the Keccak and SHA-2 extensions in addition to its RISC-V support. This brings mathematical guarantees for the correctness of these components. An even broader formal verification effort is on the roadmap, with more details coming soon.
• Continuous AI scanning: Axiom has developed a proprietary AI tool to continuously scan the OpenVM codebase as it undergoes updates and to search for any critical vulnerabilities. The tool uncovered real pre-release vulnerabilities in our internal audit process prior to external audit, and it is now an ongoing piece of our security process.

As AI-assisted attacks become ever more powerful, we believe a robust security posture for mission-critical systems like zkVMs is vital for high-stakes use cases. Our three-layered security framework, combining traditional human security review, formal verification, and AI scanning, provides enterprise customers with the highest security assurance for their workflows.

New Features

The OpenVM 2.0 production release brings GPU acceleration to Halo2 wrapping, the final step in the proving pipeline that wraps a STARK proof into a SNARK for on-chain verification on any EVM chain. This step now completes in 8.1s on a single 5090 GPU, with further performance improvements on the roadmap.

We are also introducing a new feature called deferrals, which allow an OpenVM guest program to verify proofs from arbitrary ZK circuits. This allows computation proven in a specialized circuit to be fed into general-purpose logic and included in a single OpenVM proof. At launch, deferrals have a latency cost starting at 1s for a single proof and dropping to under 300ms per proof when verifying 32 proofs.

For rollups, this enables more flexible usage of proof aggregation in the proving pipeline. A rollup typically proves its execution in chunks of blocks, which are aggregated prior to on-chain verification to reduce costs. Deferrals allow this complex aggregation logic to be expressed in an OpenVM guest program which verifies the chunk proofs, checks business logic, and produces a single proof for settlement, significantly reducing on-chain verification costs.

Deferrals also point toward future applications. For example, in verified AI inference, model inference can be proven in a custom-designed circuit tuned for optimal performance, while input pre-processing and output post-processing can be expressed as ordinary program logic in OpenVM, all within one proof. The same architecture applies to other workloads requiring custom circuits for maximal performance and composability with general-purpose computation.

We are excited to see what types of new applications deferrals can unlock. To read more about these features, please refer to the OpenVM developer docs.

Production Deployments

OpenVM 2.0 will be used to secure Lighter EVM, an EVM-equivalent rollup that can natively access the liquidity of Lighter’s perpetuals and spot markets. OpenVM 2.0’s performance and security will enable a seamless EVM development environment for developers to compose general-purpose applications against Lighter’s liquidity and markets, all while maintaining the verifiable security and performance that Lighter is known for.

Additionally, OpenVM 2.0 will continue to secure Scroll, a high-performance zkEVM Layer-2 for Ethereum. OpenVM has been finalizing Scroll mainnet since April 2025, and we look forward to the new use cases and applications unlocked on Scroll by the performance improvements of OpenVM 2.0.

Work With Us

OpenVM 2.0 is available under the MIT and Apache 2.0 licenses and is now recommended for production use. At Axiom, we know that integrating zero-knowledge proofs for your application can be a complicated task, and as a result, we provide support across the full lifecycle of ZK:

• Product design and integration: We help you understand what is possible with frontier zero-knowledge proofs, guide your choice of technical architecture to enable them, and deliver engineering engagements for custom integration.
• Support and proving services: Post-integration, we provide support including security SLAs, long-term-support (LTS) releases, and other critical services to ensure your proof pipeline can scale safely and reliably. We also optionally manage proof generation for your application, through the Axiom Proving API or dedicated on-prem deployments.

You can also refer to the developer docs to learn more about OpenVM 2.0 and see the full release on GitHub. To stay updated or get involved, join us on Telegram or follow us on X.