HiGPT is a novel heterogeneous graph language model designed to address the limitations of existing frameworks in generalizing across diverse heterogeneous graph datasets. The model introduces an in-context heterogeneous graph tokenizer that captures semantic relationships in different graphs, enabling seamless adaptation. To handle distribution shifts in relation heterogeneity, HiGPT incorporates a large corpus of heterogeneity-aware graph instructions and employs the Mixture-of-Thought (MoT) instruction augmentation paradigm to mitigate data scarcity. Extensive evaluations demonstrate that HiGPT outperforms state-of-the-art methods in both few-shot and zero-shot settings, showcasing its strong generalization capabilities and adaptability to diverse downstream tasks.HiGPT is a novel heterogeneous graph language model designed to address the limitations of existing frameworks in generalizing across diverse heterogeneous graph datasets. The model introduces an in-context heterogeneous graph tokenizer that captures semantic relationships in different graphs, enabling seamless adaptation. To handle distribution shifts in relation heterogeneity, HiGPT incorporates a large corpus of heterogeneity-aware graph instructions and employs the Mixture-of-Thought (MoT) instruction augmentation paradigm to mitigate data scarcity. Extensive evaluations demonstrate that HiGPT outperforms state-of-the-art methods in both few-shot and zero-shot settings, showcasing its strong generalization capabilities and adaptability to diverse downstream tasks.