Higgs-Llama-3-70B is a fine-tuned model based on Meta-Llama-3-70B, specifically optimized for role-playing, while maintaining competitive performance in general-purpose instruction execution and reasoning. The model is trained using supervised fine-tuning, combining annotations from human labelers and a private large language model to build preference datasets and iteratively optimize its behavior to align with system messages. Compared to other instruction-following models, the Higgs model more closely follows its assigned role.