[Aozora7]

With the release of GLM-5, I would say that they are pretty much almost as good. Basically 90% as good as Opus 4.6 on most tasks for 20% of inference cost, and open weights.