VeriScale: Adversarial Test-Suite Scaling for Verifiable Code Generation

2026-05-21

AI Summary

Existing code generation benchmarks overestimate the capabilities of large language models in adhering to specifications and producing verifiable code due to insufficient quantity and quality of positive and negative test cases. This work proposes VeriScale, a novel framework that introduces, for the first time, an adversarial implementation-driven mechanism to construct high-quality evaluation benchmarks. VeriScale employs a two-stage scaling strategy: first generating diverse and challenging test cases, then compressing them into a compact yet highly discriminative test suite. Leveraging this framework, we develop VerinaPlus and VerinaLite—two enhanced variants of the Verina benchmark. Experiments demonstrate that VerinaPlus substantially exposes deficiencies in mainstream models, leading to markedly lower task scores, while VerinaLite retains strong discriminative power with only 1/14 of the original test set size, significantly reducing evaluation overhead.

Abstract

As large language models (LLMs) are increasingly deployed for software engineering, constructing high-quality benchmarks is crucial for evaluating not just the functional correctness, but also the formal verifiability of generated code. However, existing benchmarks are limited by the quantity and quality of positive and negative test cases, leading to an overestimation of model capabilities in generating specifications and implementations. To address this, we propose VeriScale, a novel framework driven by the adversarial implementations. It consists of two stages: test-suite expansion to construct diverse and challenging test cases, and test-suite reduction to distill them into compact yet discriminative suites. While VeriScale is general, we instantiate it on Verina to construct VerinaPlus, which expands the original test suites by over 83$\times$, and VerinaLite, a lightweight 14$\times$ variant. Our experiments across eight state-of-the-art LLMs demonstrate that VerinaPlus exposes substantial model weaknesses hidden by the original benchmark, evidenced by sharp score drops on both SpecGen and CodeGen tasks, whereas VerinaLite maintains this discriminative power at a fraction of the evaluation cost. The enhanced benchmarks and source code are publicly available at https://github.com/XiaoyangLiu-sjtu/VeriScale.

Problem

• code generation
• verifiability
• benchmark
• test-suite
• large language models

Innovation

• adversarial test-suite scaling
• verifiable code generation
• test-suite expansion
• test-suite reduction
• LLM evaluation benchmark