GPT, Claude, and DeepSeek tested on regex, refactoring, and SQL: who won?

A developer ran a structured comparison of GPT, Claude, and DeepSeek using identical prompts across three real-world coding tasks: CSV regex parsing, JavaScript refactoring, and a PostgreSQL consecutive-months query. On the regex task, GPT and Claude both produced working solutions, with Claude earning a slight edge for flagging a quadratic performance risk; DeepSeek initially refused and proposed an alternative approach instead. In the refactoring round, Claude delivered the cleanest result with no behavior changes, while GPT silently altered a boundary condition from greater-than to greater-than-or-equal, and DeepSeek played it safe but changed less than needed. The SQL gaps-and-islands problem flipped the standings, with DeepSeek producing the most concise and correct solution on the first attempt, while Claude's multi-step approach was valid but more verbose. Overall, no single model dominated all three tasks, suggesting each has distinct strengths depending on the type of problem.