DetectiumFire is a newly introduced multi-modal dataset aimed at advancing the understanding of fire through the integration of vision and language data. It comprises 22,500 high-resolution images alongside 2,500 real-world fire annotations, providing a substantial resource for fire-related research. The dataset is designed to bridge existing gaps in fire data, enabling improved image generation and reasoning capabilities within the fire domain. By combining visual and textual information, DetectiumFire facilitates comprehensive analysis and modeling of fire phenomena. This dataset supports various applications, including enhanced fire detection, monitoring, and response strategies. Its release on arXiv highlights its accessibility to the research community, promoting further innovation in fire understanding. Overall, DetectiumFire represents a significant step toward more effective and data-driven approaches to fire analysis.