Windsurf Wave 14 Arena Mode: Blind A/B Model Comparison Inside the IDE

Windsurf Blog·high signal

Windsurf's Wave 14 release added Arena Mode, which runs two Cascade agents on the same prompt simultaneously with hidden model identities, letting developers vote on which output is better to discover which models actually perform best for their specific codebase. Battle Groups let you compare curated sets (e.g., fast models vs smart models) and Plan Mode adds pre-code structured planning and clarifying questions before execution. This is the first production IDE feature to generate real-world, workflow-specific model benchmarks.