I see this kind of commentary a lot but GLM-5.3 was used in July to help track down this issue. So it has been held back for a month at least. If Chinese labs are releasing LLMs w/o thoroughly testing

In fact, it seems to me Zai actually does have a much larger model in training & possibly undergoing thorough testing at the moment. 5.3 might be a final bite at the apple for a model of this size to allow it keep up w/ Moonshot while prepp — via @tphuang

High-priority alert based on source significance and keyword relevance.