“The speaker claims that both Windsor's new SWE 1.5 model and Cursor's new fast coding model are trained on Chinese foundation models.”