Air pollution is harmful to both human health and the environment, but many pollutants are difficult to monitor because ground-based stations are sparse and unevenly distributed. This study introduces a deep learning model that estimates three key pollutants—ozone (O₃), nitrogen dioxide (NO₂), and nitric oxide (NO)—at a fine spatial resolution (1 km) across the contiguous United States, every day from 2005 to 2021. These pollutants react with each other in the atmosphere, so we designed the model to estimate them together, allowing information from one to improve predictions of the others. The model uses a wide range of data sources, including weather, emissions, land cover, satellite observations, and reanalysis products. The resulting dataset will be publicly available and can support research on air quality, health disparities, and environmental policy.