PandaGPT is designed for researchers and developers to explore advanced multimodal AI capabilities. Users can leverage its open-source code, available on GitHub, to prepare and experiment with the pre-trained model. A demonstration website is also provided for direct interaction and exploration of its features. The model excels in multimodal instruction following, allowing users to issue commands across text, image/video, audio, depth, thermal, and IMU inputs. For instance, it can perform image/video grounded question answering, enabling queries about visual content. Creative writing inspired by visual stimuli is also possible, fostering new content generation methods. Its visual and auditory reasoning capabilities allow for deep analysis of multimedia data, understanding relationships and emotions. Furthermore, PandaGPT supports multimodal arithmetic, integrating numerical information from various sources for complex problem-solving. The seamless combination of semantics from diverse inputs facilitates rich compositional tasks, such as connecting visual and auditory aspects of objects. This tool is intended for research use only, serving as a foundational step towards Artificial General Intelligence. Researchers can integrate its capabilities into their projects to advance understanding in multimodal AI and large language models, contributing to the development of more holistic AI systems.