Comparison of average and median response times for models, excluding 'claude-3.7-sonnet', 'claude-3.7-sonnet-thinking', and 'auto', focusing on within-burst performance.