ResearchBenchmarksE-commerce Task Evaluation on Classic-B n=5,091 (held-out)Follow84.8B1GPT-5.584.17684.33884.584.662Jul 8, 2026Evaluation ResultsMethodMethodLinksB1B2B3B4Average Score (AVG)GPT-5.52026.0784.875.557.972.872.7Our LBMtraining_stage=base +...training_stage=base + SFT-Lotus2026.0784.297.671.17682.2