VentureBeat·Aug 6 Qwen 3.8-Max and Claude Opus 5 show why raw benchmark scores don’t predict the bill Mobile