1. Code correctness
AdequateProduces working changes that meet requirements without introducing defects.
An independent LiveSWEBench evaluation reported 43.4% average performance for Windsurf across agentic programming, targeted editing, and autocomplete tasks; its agentic-programming task resolution rate was also 43.4%. This supports competent but not consistently reliable code production, and the result is a dated snapshot using Claude 3.7 Sonnet rather than evidence of current performance across models.
Evidence confidence: medium