If Beam "rivals GLM 5.2 on reasoning" does it mean it's as good as GLM 5.3 Flash? (a much smaller model)