Qwen 3.8-Max and Claude Opus 5 show why raw benchmark scores don't predict the bill

Alibaba released Qwen 3.8-Max this week and marketed the preview as second only to Claude Fable 5 (their launch-day table was more equivocal: the model leads on one of 12 coding-agent rows). But an independent harness came close to the oppo