“The open source model GLM 5.2 is performing at a very high level, to the point where half of Factory's internal token usage is now on open models.”