Pico Prism: One Stone, Two Wins — Speed and Efficiency

Key Takeaways

• To transition from re-execution to ZK verification at L1, Ethereum has set performance benchmarks collectively known as Real-Time Proving (RTP): proof generation within 10 seconds and hardware costs under $100K.

• In Oct 2025, Brevis’s Pico Prism demonstrated the technical feasibility of RTP by achieving 96.8% proof coverage within 10 seconds on 45M gas limit blocks — without the need for massive hardware resources, at a cost of approximately $128K.

• This result was achieved by extreme hardware parallelization through a modular zkVM architecture and app-level coprocessor design optimized for L1 EVM execution, unlike general-purpose VMs.

• Pico Prism proved state-of-the-art (SOTA) performance, bringing Ethereum's L1 zkEVM vision within reach. Separately, Brevis's ZK technology already enables applications like verifiable AI and intelligent DeFi through its ZK Data Coprocessor. For Pico Prism to become core Ethereum infrastructure, it must still address challenges such as further cost reduction (toward home proving), open-sourcing, and production-level security reliability.

1. Ethereum’s Dream: Transitioning to ZK Verification

1.1 Real-Time Proving

To realize its vision of becoming a “World Computer,” Ethereum has long grappled with scalability challenges. Although it has introduced various L2 solutions such as rollups to improve throughput, the Ethereum mainnet still faces a fundamental bottleneck — every node must re-execute and verify all transactions.

To overcome this limitation and move toward the "Gigagas" era, Ethereum is pursuing a transition to L1 zkEVM — an architecture where validators verify cryptographic proofs instead of re-executing transactions. Instead of every validator independently re-executing an entire Ethereum block, zkVM (zero-knowledge virtual machine) and zkEVM (zero-knowledge Ethereum virtual machine) technologies can prove that "all operations in this block were executed correctly."

Real-Time Proving (RTP) is the performance benchmark that zkVMs must meet to make this transition feasible: generating zero-knowledge proofs within 10 seconds while meeting hardware cost and power constraints. The Ethereum Foundation set these targets to ensure that proof generation is fast enough to maintain network liveness and cheap enough to preserve decentralization.

Once Ethereum transitions to ZK verification at L1, the network's efficiency and throughput will significantly improve — nodes will only need to verify compact proofs instead of re-executing every transaction. This will enable higher gas limits, reduced congestion, lower costs, and broad benefits for both the L2 ecosystem and end users.

1.2 Two Pillars of RTP: “Speed” and “Economic Accessibility”

To ensure the smooth adoption of L1 zkEVM, a change that promises to benefit everyone, the Ethereum Foundation announced in a July 2025 blog post four specific benchmarks for what t:

1. Proof Coverage: More than 99% of all Ethereum blocks must be proven stably while meeting the “proof latency” criteria below.

2. Proof Latency: Proof generation must complete within 10 seconds.

3. Hardware CAPEX: The total hardware investment required for proof generation must be under $100,000.

4. Power Draw: Proof generation in an on-premise (locally owned and operated) environment, not via cloud or third-party services, must consume less than 10kW of power.

What’s particularly notable here is that Ethereum’s criteria emphasize not only speed (criteria 1 and 2) but also economic accessibility (criteria 3 and 4). The inclusion of economic accessibility underscores that Ethereum’s goal is a decentralized prover network where solo stakers running validators from home can also run provers at reasonable cost, what the Foundation calls "home proving.

Even if a million-dollar supercomputer could generate proofs in five seconds, such centralization of proving power would contradict Ethereum’s philosophy. Therefore, economic accessibility is both a technical and a philosophical challenge for RTP.

How difficult it is to achieve both “speed” and “economic accessibility” simultaneously can be seen in the case of Succinct’s SP1 Hypercube, which represented the state-of-the-art (SOTA) performance as of May 2025.

Built on Succinct’s high-performance SP1 zkVM, the SP1 Hypercube was designed to parallelize proof computations across a large-scale GPU cluster environment.

At the time, SP1 conducted its tests on Ethereum mainnet blocks with a gas limit of 36M, under the condition of generating proofs within 12 seconds. As a result, it achieved 92.8% proof coverage within 12 seconds using a cluster valued at approximately $256K (consisting of 160 NVIDIA RTX 4090 GPUs), demonstrating the technical feasibility of real-time proving. However, these results were based on a “within 12 seconds” benchmark, not the EF’s target of “99% coverage within 10 secs”, and still fell significantly short of the “under $100K cost” criterion.

2. Pico Prism: Half Cost, Faster Speed

2.1 Efficiency Proven by Numbers

Until mid-2025, the competition was essentially a hardware arms race aimed at achieving the goal of “proofs within 10 seconds.” However, the SP1 case (approximately $256K cost, 92.8% coverage within 12 seconds) failed to meet either objective.

What completely upended this landscape was the debut of Brevis’s Pico Prism on Oct 15, 2025. When tested on the same 36M gas limit blocks that SP1 used, Pico Prism achieved an average proving time of 6.04 secs and 98.9% coverage within 10 secs. It also set a new benchmark on larger 45M gas limit blocks, with an average proving time of 6.9 secs and 96.8% coverage within 10 secs.

These results represented the closest achievement yet to the EF’s definition of RTP — proof generation within 10 secs using hardware costing under $100K. Following the announcement, it received public endorsement from Vitalik Buterin, establishing itself as the new SOTA in the field.

As shown in the table below, under the 36M gas limit block condition, Pico Prism reduced hardware costs by 50% compared to SP1 Hypercube, improved sub-10s proof coverage by 2.5×, and achieved an average proving speed 71% faster — resulting in an overall 3.41× improvement in performance efficiency.

Most importantly, this achievement demonstrated that RTP is no longer the exclusive domain of enterprise-scale data centers. While the race up to Q2 2025 was about raw hardware capacity, the competition after Pico Prism shifted its focus toward efficiency.

2.2 Pico and Pico Prism: From General-Purpose Engine to Specialized Cluster

To understand Pico Prism, we must first look at its foundation — Brevis’s Pico zkVM. Pico is a general-purpose zkVM designed to execute arbitrary Rust-based programs off-chain and verify their results on-chain through ZKPs. However, as a general-purpose design, Pico faced limitations in handling the high-intensity computation and ultra-fast, large-scale parallel processing required for RTP-level performance — proving a 45M gas block within 10 seconds.

To overcome this limitation, Brevis introduced Pico Prism on Oct 15, 2025. Pico Prism inherits Pico’s modular architecture but re-engineers it into a specialized distributed multi-machine cluster tailored specifically for RTP.

Whereas the original Pico handled all proving operations within a single system, Pico Prism extends this process across multiple servers and GPUs through distributed parallel computation.

Thanks to this architecture, Pico Prism successfully combines the flexibility of a general-purpose zkVM with the performance of a specialized system, bringing the EF's RTP performance targets within practical reach.

2.3 Secret of Efficiency: Modular zkVM Architecture

How was Pico able to achieve such high efficiency? The answer lies in its modular architecture.

Most traditional zkVMs, such as SP1, are built on general-purpose instruction sets (like RISC-V), using a monolithic structure where a single large VM handles all computations. While this design offers strong generality, it is difficult to apply task-specific optimizations (such as those required for RTP), and it lacks architectural flexibility.

In contrast, Pico is designed as a modular zkVM, much like building with LEGO blocks, allowing developers to assemble and optimize proof system components according to application needs. For example, developers can directly choose or fine-tune proving backends and proving workflows, balancing security and performance to create a customized proving environment.

Pico Prism, built on this modular foundation, introduces a multi-machine cluster architecture designed specifically for the demanding workload of RTP. The entire proving pipeline was restructured into two main stages:

1. Emulation stage: Simulates block execution to generate operation traces.

2. Layered recursion stage: Compresses these traces through multiple recursive layers to produce the final proof.

Throughout this process, Pico Prism precisely distributes workloads between CPUs and GPUs so that dozens of GPUs operate at maximum efficiency without idle time (fully saturated). In other words, it builds an “embarrassingly parallel” pipeline that maximizes hardware utilization.

This approach enabled a level of computational density that was practically impossible in monolithic zkVMs, making Pico Prism the pivotal breakthrough that opened the RTP era.

* RISC-V is a royalty-free, open instruction set architecture (ISA) that allows modular, purpose-optimized chip design.

2.4 Maximizing Efficiency: App-Level Coprocessor

The key to Pico Prism’s exceptional efficiency, built on an architecture already optimized for performance, lies in its “coprocessor” strategy, which amplifies the strengths of its modular design. This approach draws inspiration from Vitalik Buterin’s “Glue-and-Coprocessor” architecture and extends it even further.

In this architecture, the on-chain layer (such as the EVM) acts as the “Glue,” responsible for coordination and verification, while the off-chain modules serve as “Coprocessors,” performing computationally intensive tasks, such as generating zero-knowledge proofs, and then returning cryptographic proofs of their results to the glue layer for validation. It’s a modular, efficiency-centered model that separates lightweight control from heavy computation.

The efficiency of this Glue-and-Coprocessor model largely depends on which level it is implemented at. Most zkVMs, including many based on RISC-V ISAs, adopt this concept at the function level, by adding “precompile” layers — effectively function-level coprocessors. For instance, a precompile might accelerate expensive yet frequently used operations like Keccak hashing by handling them through a dedicated processor.

However, Pico Prism pushes this idea much further with “App-Level Coprocessors” — a defining feature of its architecture. Instead of merely accelerating individual functions (e.g., hash operations), Pico embeds ZK coprocessors optimized for entire high-level application tasks, such as on-chain DA or ML/AI computation.

By focusing optimization not just at the function layer but at the application layer, Pico Prism achieves far greater computational efficiency. Its outstanding performance in RTP stems precisely from this design: an extremely optimized, modular, and distributed-parallel architecture purpose-built for a specific workload — L1 EVM execution.

3. The Future Pico Prism Will Bring

3.1 Already-Begun Era of “Verifiable Applications”

The future that Pico Prism will bring is not a distant story. Independent of Ethereum’s L1 zkEVM roadmap, Brevis’s technology already enables verifiable computations for applications, realizing functions that were impossible through smart contracts alone.

• Intelligent DeFi: Complex trading strategies, risk analysis, and user reputation–based dynamic interest rates become verifiable on-chain through Brevis. For example, PancakeSwap offers a trading volume–based fee discount feature by proving past transaction data off-chain and submitting a ZKP.

• Verifiable AI: AI agents can analyze on-chain data and submit their conclusions (inferences) together with ZKPs. This forms the foundation of a “trustworthy AI” economy and has already been implemented through Brevis’s ZK coprocessor technology.

• Cross-Chain Data Verification: Brevis allows smart contracts to verify and access data from other blockchains via ZK proofs, enabling cross-chain applications without relying on multisig bridges.

3.2 Ethereum’s Leap Forward

Building on the technological advancements already being realized, the future that Pico Prism seeks to accelerate is Ethereum’s transformation into a real-time verifiable L1 zkEVM. This monumental paradigm shift will evolve Ethereum beyond a mere L1 into a “global verification layer” grounded in mathematical certainty. Once Ethereum can verify every block in real time, the protocol will undergo the following fundamental changes:

• Maximizing the Economics of L2s and Rollups: One of the largest costs for current ZK rollups is the gas fee required to verify proofs on Ethereum. If RTP becomes possible on Ethereum, the gas fee ZK rollups pay for proof verification will converge toward zero. This will dramatically lower L2 transaction costs, opening the door to mass adoption and accelerating the rise of numerous app-chains.

• Higher Gas Limits and Throughput: When Ethereum validators verify proofs instead of re-executing transactions, the network can support much higher block gas limits. This means more transactions per block without increasing hardware requirements for validators, resulting in significantly improved scalability for Ethereum.

3.3 From Technical Leader to Core Ethereum Infrastructure

As Pico Prism, which has already achieved an undisputed State-of-the-Art (SOTA) status in the RTP field, leads the adoption of Ethereum’s L1 zkEVM, the vision of a “verifiable internet” will become a reality, and the demand for proof generation will surge explosively.

However, having a superior position in benchmarks does not automatically guarantee a place as Ethereum’s verification engine. For Pico Prism to establish itself as Ethereum’s public verification infrastructure, it must overcome not only technical challenges but also the greater task of proving its “trust” within the ecosystem.

• Achieving Cost Efficiency (Home Proving): Currently, Pico Prism achieves the Ethereum Foundation’s target performance level of around $100K using 64 high-end GPUs (approximately $128K in cost). However, this is still far from true “home proving.” Brevis’s next roadmap — to achieve 99% RTP with fewer than 16 GPUs (approximately $40K in cost) — goes beyond simple cost reduction, it represents the realization of genuine home proving.

Source: ethproofs.org

• Ensuring Accessibility: As highlighted by the Ethproofs dashboard, Pico Prism’s prover is not yet open source, raising concerns about decentralization and accessibility. To address these “vendor lock-in” fears, Brevis has begun aligning with EF standards, open-sourcing Pico’s modular architecture and releasing its SDK since November 2025, but open-sourcing Pico Prism itself remains a critical next step.

• Production-Grade Security and Reliability: A bug in any model adopted as Ethereum’s verification engine could lead to catastrophic consequences. Therefore, production-level security and reliability are non-negotiable. Before any L1 integration, Pico Prism must undergo rigorous audits and prove stability under real-world production conditions, surpassing mere benchmarking success.

In summary, Pico Prism has already proven that the EF's RTP benchmarks are technically achievable. Now, it faces a deeper question: “Can its efficiency and scalability endure in a truly decentralized environment?”

If Pico Prism can extend its technical leadership into open-source transparency and protocol-grade security, it will transcend being merely a cutting-edge zkVM — becoming core infrastructure for Ethereum's L1 zkEVM vision”

That is why the next chapter in Pico Prism’s evolution is so crucial — not just for Brevis, but for Ethereum’s future itself.