Tomorrow.io operates a near-global, near-real time precipitation retrieval framework that extends coverage beyond traditional radar networks. This manuscript details the development and operationalization of the product's third generation, which is currently available with a refresh rate of 5 minutes and spatial resolution of 0.04° (~4 km), covering latitudes between 60°S and 72°N. Precipitation estimates are derived from a convolutional neural network that ingests observation sequences from multi-channel geostationary imagers and orbiting microwave sounders within the preceding hour. The network is trained to handle sparse data from individual instruments, delivering a spatially-complete retrieval that approximates real-time conditions. Our results demonstrate that this third generation of the Tomorrow.io near-global precipitation product improves categorical and continuous precipitation metrics by 5-10% over the previous generation, with significantly larger skill improvements over publicly available near real-time alternatives such as IMERG-Early. Verification was performed against terrestrial radar networks (USA, Europe, Japan, Brazil, and Australia) using test datasets spanning all seasons and against gauge observation networks for several case studies outside the training domains. Additionally, case studies of retrievals in impactful hurricanes during 2024 and recent operational results demonstrate the impact of incorporating microwave sounder data from two recently launched instruments in Tomorrow.io's constellation. These instruments provide more direct observations of precipitation processes than the geostationary imagers, which the network leverages to further enhance skill beyond the third-generation product's elevated baseline. The Tomorrow.io satellite constellation and precipitation retrieval framework benefit near-real time weather applications in aviation, agriculture, hazard preparedness, water resource management, and other sectors at global scales.