Building age information provides an important basis for understanding urban spatial evolution, historic landscape conservation, and existing building stock renewal. To address the high cost and long cycle of traditional age identification, as well as the limited stability of single-model recognition in complex street-view environments, this study takes Harbin as the research area and develops a building age recognition method based on multi-model fusion. Building footprints from OpenStreetMap are used as the spatial positioning basis, while street-view images and user-generated content images are integrated and screened through semantic segmentation. Three deep learning models are trained for age classification, and their outputs are further integrated through voting fusion, weighted fusion, and stacking fusion. Model performance is evaluated using confusion matrices, category-level metrics, and feature projection analysis. The results show that all three single models achieve good age discrimination performance, while stacking fusion performs best and further improves recognition accuracy and category separability. City-scale recognition results indicate that buildings constructed after 2000 constitute the main body of Harbin’s building stock, while early-period buildings account for a relatively small proportion. Spatially, buildings of different periods show a gradual expansion pattern from the historic urban core to the periphery. This study can provide methodological reference and data support for urban building stock surveys, historic landscape conservation, and renewal zoning decisions.