Community builds with refusal behaviour edited out of open weights. The articles explain how the edit works and what each build costs.
Two weight-edited GLM-5.3 derivatives that refuse far less than the base models. What abliteration changes, which base each starts from, what each costs, and the limits that still apply.