Cities are major sources of greenhouse gas emissions, and combining activity-based inventories with atmospheric constraints can support science-based mitigation at subnational levels. We implement high-resolution WRF-Chem simulations over the Greater Tokyo Area (GTA) using five anthropogenic emission inventories (MOSAIC, ODIAC, EDGAR, CAMS-GLOB-ANT, and GCP-GridFED) and examine CO₂ concentrations at four sites. Using observations from the summit of Mt. Fuji as a background reference, we derive GTA CO₂ enhancements (ΔCO₂). The simulations reproduce most observed ΔCO₂ variability and the relationship between elevated ΔCO₂ and transport from the urban core and the Tokyo Bay, while showing differences in ΔCO₂ amplitudes due to five inventories. The four observation sites represent different GTA subregions, providing complementary constraints on inventory spatial patterns and highlighting the importance of site selection in urban assessments. High simulated ΔCO₂ at Tsukuba is largely attributable to stronger northern suburban emissions in MOSAIC and ODIAC, whereas high ΔCO₂ at Yoyogi in ODIAC and GridFED is attributable to stronger nearby urban-core emissions. CAMS-GLOB-ANT and GridFED overestimate ΔCO₂ at Yokosuka because of large emissions along the Tokyo Bay. EDGAR shows the lowest bias, although its spatial distribution is coarse and diffuse compared with MOSAIC and ODIAC. Anthropogenic CO₂ is best reproduced by EDGAR and MOSAIC, though this result is strongly influenced by Yoyogi. Corrected total GTA CO₂ emissions of inventories range from 60.2 to 81.1 Tg C yr^−1. These results show where inventories diverge within a megacity and provide a practical benchmark for improving urban CO₂ inventories and future urban emission monitoring.