The Android Multi-annotation EXpo (AMEX) dataset is introduced to advance research on mobile GUI agents. AMEX provides a comprehensive, large-scale dataset with three levels of annotations: (i) GUI interactive element grounding, (ii) GUI screen and element functionality descriptions, and (iii) instructions with GUI-action chains. It includes over 104K high-resolution screenshots, 21K screen descriptions, and approximately 3,000 complex instructions. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX includes three levels of annotations: (i) GUI interactive element grounding, (ii) GUI screen and element descriptions, and (iii) instructions with GUI-action chains. The dataset comprises over 104K high-resolution screenshots, 21K screen descriptions with 300K element-wise functionalities, and approximately 3,000 unique complex instructions, with an average of 12.8 steps. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset isThe Android Multi-annotation EXpo (AMEX) dataset is introduced to advance research on mobile GUI agents. AMEX provides a comprehensive, large-scale dataset with three levels of annotations: (i) GUI interactive element grounding, (ii) GUI screen and element functionality descriptions, and (iii) instructions with GUI-action chains. It includes over 104K high-resolution screenshots, 21K screen descriptions, and approximately 3,000 complex instructions. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX includes three levels of annotations: (i) GUI interactive element grounding, (ii) GUI screen and element descriptions, and (iii) instructions with GUI-action chains. The dataset comprises over 104K high-resolution screenshots, 21K screen descriptions with 300K element-wise functionalities, and approximately 3,000 unique complex instructions, with an average of 12.8 steps. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is annotated by human annotators trained with detailed guidelines to ensure precision and quality. AMEX is designed to provide a multi-level understanding of mobile GUIs, enabling more effective GUI agents. The dataset is