FibRace: Mobile ZKP Benchmark

• FibRace is a large-scale empirical benchmark that evaluates mobile client-side zero-knowledge proof generation using the Cairo M ZK-VM.
• The initiative uses a gamified Fibonacci proof challenge to measure performance metrics such as latency, proving frequency, and hardware dependency across thousands of devices.
• The collected dataset provides practical insights for optimizing ZKP systems and scaling onchain verification for privacy-preserving mobile applications.

FibRace is a large-scale empirical benchmark designed to evaluate the feasibility and performance of client-side zero-knowledge proof generation on mobile devices using the Cairo M proof system. Developed by KKRT Labs and Hyli, the project was implemented as a gamified mobile application—inviting users to generate proofs of Fibonacci numbers and submit them for onchain verification. The experiment simultaneously targeted public engagement and provided the most comprehensive measurement to date of mobile proving capabilities, systematically cataloging performance across thousands of devices and millions of proofs (Malatrait et al., 16 Oct 2025).

1. Objectives and Experimental Design

FibRace was conceived as a dual-purpose initiative: (1) engaging the public through a competitive mobile game built around mathematical proof generation, and (2) executing a rigorous empirical evaluation of client-side proof computation using Cairo M, a ZK-VM (Zero-Knowledge Virtual Machine) optimized for resource-constrained hardware.

• Gamification: The mobile app randomly selected an index $n$ in the Fibonacci sequence (range: $1 \leq n \leq 100\,000$). Players were tasked to compute $F_n$ and simultaneously generate a zero-knowledge proof attesting correctness. The output was a collector card cataloging the value and proving time, with leaderboards ranking users by proof speed, frequency, and overall progression.
• Proof Generation: Computation and proof generation occurred locally. The Fibonacci circuit definition adhered to $F_n = F_{n-1} + F_{n-2}$ with appropriate initial conditions, embedded in the M31 prime field and affording a proof security level of 96 bits.
• Data Collection: For each proof attempt, the client app recorded device metadata (model, OS version, RAM, SoC) and proof execution statistics. Over the three-week experiment (September 11–30, 2025), 6,047 participants generated 2,195,488 proofs on 1,420 unique smartphone models.

This setup allowed FibRace to aggregate device-level data on proof latency, reliability, and hardware characteristics with high statistical power.

2. Zero-Knowledge Proof Methodology and Workflow

• Cairo M Integration: Cairo M, a mobile-optimized ZK-VM, was embedded in the application to perform both the Fibonacci computation and the ZKP generation. The circuit for Fibonacci calculation was expressed as a sequence of field additions, concluding with a ZKP that could be validated externally.
• Client-side Proving: All proof computation was executed locally on the device, without recourse to remote outsourcing or specialized hardware. This architecture safeguards privacy and provides realistic performance statistics absent of server-side optimizations.
• Onchain Verification: Once generated, proofs and collector cards were submitted to Hyli's blockchain, which natively validates Cairo M proofs using its Autobahn consensus driver. The protocol is designed to absorb high submission volumes—peak observed throughput was 260,000 proofs/hour—with no observable network congestion.

3. Performance Analysis: Device, RAM, and SoC Correlates

The experiment produced highly granular datasets characterizing proof generation under varying device constraints.

• Latency: Median proof generation time was 6.37 seconds; 90th percentile latency was 13.75 seconds. Most recent smartphones completed proofs in under 5 seconds, aligning with target user experience constraints.
• Proving Frequency: Mean device proving frequency was 30kHz (Cairo M cycles/sec); median value, affected by slower devices, was 10kHz.
• RAM Dependency: Devices required a minimum of 3 GB RAM for reliable proof computation. Below this threshold, crash rates increased sharply—documented at 0.72% overall, mostly in the 3–4 GB RAM bracket. Stability improvements plateaued above 6 GB RAM.
• SoC Benchmarking: Apple’s A19 Pro and M-series were the fastest, with $<$3s proving times at high cycle rates. The upper echelon of the performance table was dominated by Apple chips, followed by top Qualcomm, MediaTek, Kirin, and Exynos models.
• Device Model Distribution: The dataset included 1,420 device models, spanning flagship and budget categories, providing hardware-stratified insights into ZK performance.

Table: Summary of Key Proving Performance Metrics

Metric	Average	Median	90th %ile	RAM (min)	Crash Rate
Proving time (sec)	<5	6.37	13.75	3 GB	0.72%
Proving freq (kHz)	30	10	N/A	6 GB+ optimal	--

4. Blockchain Verification and Scalability

Hyli's blockchain recorded all proofs submitted from the mobile clients and performed native onchain verification using its Autobahn consensus protocol.

• Scalability: The platform demonstrated the ability to handle peak submissions ($\sim$260k proofs/hour) without congestion or verification bottlenecks.
• Protocol Integration: Proof verification is performed in Cairo M; the pipeline ensures that collector cards (proof + metadata) are cryptographically authenticated and persisted onchain for leaderboard and audit purposes.

This architectural design validates the proposition that distributed, privacy-preserving computation with immediate cryptographic attestation can be realized at planetary scale with no infrastructural bottlenecks.

5. Implications for Zero-Knowledge Proof Engineering

The results from FibRace substantiate several important points for ZKP research and mobile application development:

• Client-Side ZK Feasibility: Modern smartphones, once equipped with adequate memory and SoC performance, are capable of directly producing ZK proofs—inverse computation workloads previously reserved for remote servers can now be shifted to clients.
• Infrastructure Simplification: Eliminating remote provers reduces system complexity, mitigates privacy risk, and enhances user trust.
• Baseline for Optimization: The rich dataset (millions of real proofs, device statistics) provides a practical baseline against which future Cairo M and general ZKP performance enhancements can be measured.
• Application Opportunities: The proven model supports new classes of privacy-preserving mobile workflows—identities, attestations, in-app game mechanics, etc.—all verified locally and cryptographically secured for decentralized use.

A plausible implication is that continued advances in lightweight proving schemes and mobile hardware may bring multi-purpose, proof-powered infrastructure to the mainstream, including offline and interactive privacy applications.

6. Future Directions

FibRace suggests several concrete avenues for future research:

• Circuit Complexity: Extending benchmarks beyond Fibonacci—incorporating cryptographic primitives (e.g., SHA2) and real-world transaction logic to stress-test ZK-VMs on mobile.
• Power and Thermal Metrics: Quantifying the energy footprint and heat generation during sustained proving sessions, enabling optimization for battery-constrained environments.
• Software Robustness: Investigating rare crash scenarios (sometimes observed even at high RAM) to further minimize client-side instability—thread management and resource allocation improvements are required.
• Gamified Benchmarks for Other Protocols: Generalizing the FibRace methodology to evaluate alternative ZKP systems, accelerating the practical feedback cycle between cryptographic engineering and end-user experience.

7. Conclusion

FibRace demonstrates that client-side zero-knowledge proof generation with Cairo M is technically feasible, secure, and performant at scale on modern smartphones. RAM and SoC characteristics dominate execution profiles, but onchain integration (via Hyli’s blockchain) resolves all proofs at high volume. The resulting dataset sets an empirical baseline for future ZKP developments and initiates a new phase in mobile privacy infrastructure, where rigorous cryptographic operations can be performed universally at the client layer. These findings catalyze research into lightweight provers, decentralized identity, and privacy-preserving protocols for pervasive mobile environments (Malatrait et al., 16 Oct 2025).