SCDBench: A Benchmark for LLM-Based Smart Contract Decompilers

Researchers introduced SCDBench, a comprehensive benchmark dataset with 600 real-world Solidity contracts designed to rigorously evaluate LLM-based smart contract decompilers. Testing frontier models like Claude Opus and GPT-5.3-Codex revealed significant limitations: the best-performing model achieved semantic consistency on only 42/600 contracts, highlighting that while LLMs can generate compilable code, accurately recovering original contract semantics remains an unsolved challenge critical for blockchain security.

Smart contract decompilation—the process of reconstructing readable source code from blockchain bytecode—has become increasingly important as the industry seeks transparency and security auditing mechanisms. SCDBench addresses a critical gap in evaluation methodology by establishing standardized metrics and a substantial dataset that moves beyond narrow testing approaches. The benchmark's four-stage evaluation framework (format completeness, compilability, ABI recovery, and semantic consistency via differential replay) provides a more comprehensive assessment than existing methods, which often rely on inconsistent metrics that obscure actual decompiler reliability.

The research emerges at a pivotal moment when LLMs demonstrate remarkable capability in code generation yet simultaneously risk producing plausible-looking but semantically incorrect output. The finding that frontier models achieve only 42/600 perfect decompilations exposes a critical weakness in current approaches, particularly concerning given the security implications of flawed contract reconstruction. This limitation directly impacts blockchain auditors, developers, and security researchers who depend on decompilers to analyze unverified contracts and identify vulnerabilities.

For the cryptocurrency and security sectors, SCDBench establishes a foundation for measurable progress in decompiler development. The introduction of same-model compilation-repair shows promising improvements with reasonable cost increases, suggesting a pathway toward more reliable tools. However, the persistent semantic consistency gap indicates that fully autonomous decompilation remains impractical for security-critical applications, necessitating continued human oversight during contract analysis and highlighting the market demand for improved decompilation infrastructure.

• →SCDBench's 600-contract dataset with replayable semantic checkpoints enables rigorous, reproducible evaluation of smart contract decompilers.
• →Best-performing frontier LLMs achieve semantic consistency on only 7% of test contracts, revealing significant limitations in current decompilation approaches.
• →Four-stage evaluation framework addresses critical gaps by assessing format, compilability, ABI recovery, and semantic consistency rather than using inconsistent metrics.
• →Compilation-repair augmentation substantially improves decompiler performance at modest computational cost, suggesting a practical near-term improvement pathway.
• →Persistent semantic gaps indicate fully autonomous decompilation remains impractical for security-critical blockchain applications requiring human oversight.