Ant Group's Bailing Move Is More Ruthless Than DeepSeek

Late last Friday night, a friend working as an architect at Alibaba Cloud sent me a screenshot, saying the internal discussion about Ling-3.0-Flash had exploded within the team. 124B parameters, 5.1B active, inference costs plummeted by 70%. I asked him, isn't this just a variant of MoE? He said, no, this is native hybrid reasoning, not that routing gating stuff.