Zero-knowledge proofs (ZK proofs) have crossed a critical threshold in 2026, shifting from theoretical cryptography to practical infrastructure that powers identity verification, blockchain scaling, and compliance systems. The cost to prove an Ethereum block fell from minutes on specialized hardware in 2024 to under 12 seconds on consumer GPUs by late 2025, a roughly 45-fold collapse in proving expenses that fundamentally changed the question from "can we afford this?" to "what is worth proving?"

What Changed Between 2024 and 2026?

Three major shifts reshaped the zero-knowledge proof landscape. First, specialized circuit design languages gave way to general-purpose virtual machines (zkVMs) that let developers write ordinary Rust code, compile it to a standard RISC-V target, and automatically generate proofs of correct execution. This eliminated the months of specialist work previously required for each new feature.

Second, the Ethereum Foundation set concrete proving targets in July 2025: blocks proven in under 10 seconds on a 100,000-dollar rig drawing at most 10 kilowatts, with proofs smaller than 300 kilobytes and no trusted setup required. By the end of 2025, multiple teams hit those targets. The public ethproofs tracker showed average block-proving costs plummeted from 1.69 dollars in January to under 4 cents in December 2025.

Third, major technology companies validated non-blockchain use cases. Google announced zero-knowledge age assurance for its Wallet in April 2025 and open-sourced the underlying Longfellow library in July 2025, signaling serious investment in identity and privacy applications beyond cryptocurrency.

Which Zero-Knowledge Proof Systems Are Production-Ready?

The field narrowed to a conservative shortlist of mature systems, though the choice depends on specific workload requirements. A comparative benchmark published by Fenbushi Capital in August 2025 evaluated eight zkVM systems and found the most robust performers showed near-constant memory use, while several younger systems experienced sharp increases in proving time and memory as input size grew.

• SP1 (Hypercute): Proved 99.7 percent of Ethereum blocks in under 12 seconds on 16 RTX 5090 GPUs as of November 2025. Currently live on mainnet infrastructure and considered production-ready.
• RISC Zero: Demonstrates mature GPU proving with graceful scaling in independent benchmarks. Powers live coprocessors and is in production use.
• OpenVM 2.0: Released as a production system in July 2026 after external audit. Reported 9.8-second P99 Ethereum proving on eight RTX 5090 GPUs, meeting the Ethereum Foundation's latency targets.
• Airbender (Matter Labs): Integrated into the ZKsync OS developer-preview testnet but designed specifically for that ecosystem, not as a generic production zkVM. Externally assessed but not recommended for general-purpose applications.
• a16z crypto system: Achieved over 1 million RISC-V cycles per second on a 32-core CPU and roughly 50-kilobyte proofs after the Twist and Shout upgrade in 2025. Remains in alpha and explicitly not recommended for production.

How to Evaluate a Zero-Knowledge Proof System for Your Use Case

• Benchmark your exact program: Pin an audited release and run your specific workload on the target hardware before committing. Proving costs vary dramatically based on circuit complexity, wrapper design, verifier type, and hardware availability, so generic benchmarks do not predict your application's performance.
• Measure latency requirements: Real-time proving of Ethereum blocks takes 10 to 12 seconds on 16 consumer GPUs for blocks with hundreds of transactions. Smaller programs prove in seconds on a single GPU. Verify these timelines match your infrastructure and business requirements.
• Account for proof size and verification cost: Modern STARK-based systems wrap proofs in a final Groth16 or PLONK proof for cheap on-chain verification, landing in the tens-of-kilobyte to sub-kilobyte range. However, that final wrap reintroduces pairing-based cryptography, so post-quantum resistance requirements may force you to accept larger raw STARK proofs.
• Assess security and audit status: Prefer systems with published external audits and clear documentation of known limitations. Younger zkVMs may offer novel features but lack the battle-testing of mature alternatives.

Where Are Zero-Knowledge Proofs Actually Being Used Today?

Beyond blockchain scaling, zero-knowledge proofs enable practical applications that were impractical or impossible before the cost collapse. Mobile proving through the Mopro toolkit wraps existing provers for iOS and Android, though benchmarks show multi-second proof times and verification above 100 milliseconds depending on the circuit.

zkEmail lets users prove facts about DKIM-signed emails, such as receiving a termination letter from an employer, without revealing the full email. This is powerful for privacy-preserving credential verification, though the regex component had eleven confirmed constraint bugs found through fuzzing in 2025, highlighting the importance of security audits.

zkTLS for AI agents proves facts about HTTPS sessions without requiring websites to integrate new attestation APIs. The design space has evolved beyond a clean split between multiparty computation (MPC) and proxy approaches. TLSNotary now offers MPC-TLS as its default with a faster proxy mode available for different threat models.

Identity and personhood verification emerged as the most mature category. Projects like Self, Rarimo, and Anon Aadhaar prove age, nationality, or uniqueness from government documents via NFC chips or signed QR codes. World ID deployed millions of verified users on Semaphore-based proofs, moving from demo to production in roughly 18 months.

Regulatory timing accelerated adoption. Every European Union member state must offer a European Digital Identity Wallet by the end of 2026 under eIDAS 2.0, and the framework explicitly favors selective disclosure. Google's open-sourced Longfellow ZK library targets this regulatory window, with age assurance as the first use case.

The 45-fold cost collapse in one year fundamentally reframed the economics of zero-knowledge proofs. The question is no longer whether proving is affordable, but which applications justify the engineering investment and infrastructure costs. For architects evaluating zkVMs, the path forward requires benchmarking exact workloads, understanding the security audit status of chosen systems, and aligning proof latency and size with business requirements rather than relying on vendor claims or generic performance metrics.