We develop a global dataset of daily fire-sourced PM_2.5 concentrations at a spatial resolution of 0.25° for 2000–2023, using a chemical transport model driven with two fire emission inventories and a machine learning approach trained with ground measurements from over 9000 sites. The dataset shows significant spatiotemporal variations of fire PM_2.5 in the past decades, serving a useful tool for exploring the impacts of fire-related air pollutants on climate, ecosystems, and public health.