Qwen2.5-Coder vs DeepSeek-Coder for Solidity Review: What I Actually See Locally

Short summary

A developer compares Qwen2.5-Coder and DeepSeek-Coder (7B and below) for reviewing ten planted-bug Solidity contracts via Ollama on WSL2. Qwen wins on instruction following and format discipline, making it better for automated pipelines, while DeepSeek shows stronger raw pattern recall and skepticism but produces more noise. Neither 7B model is an auditor replacement, but both are useful assistants on digestible code.

• •Qwen2.5-Coder excels at structured output format compliance, critical for pipeline integration
• •DeepSeek-Coder has better raw pattern recall on classic Solidity bugs but more false positives
• •Both models tested on 10 small contracts with planted bugs; neither suitable for large real-world audits

Generated with AI, which can make mistakes.