A user reports that the 3.8-27B model has made previous 3.5/3.6-35B models obsolete, describing it as "absurdly superior" for applied science workflows including data pipeline design, analysis, and report writing. While the 27B model requires 3–4x more wall-clock time to complete tasks, the quality and attention to detail are dramatically higher. The author, who has access to the Z.ai API, compares its output against models 5.3 and 5.3-flash.