Each search provider is benchmarked the way an AI agent uses it — its results feed a fixed answer model (GPT-5.6 Luna), and the answers are graded across five public QA benchmarks. Every You.com sampler is highlighted; lower time and cost, and higher accuracy, is better.