GPT-5.4 leads in lines, Claude Sonnet 4.6 in security issues

Sonar evaluated 4,444 Java assignments across 53 LLMs. GPT-5.4 produced 1.2M lines of code, while GPT-4o generated under 250K. Claude Sonnet 4.6 had the highest security issue rate at 300 per million lines.