<p>To curb the spread of COVID-19 pandemic, many countries around the world imposed an unprecedented lockdown producing reductions in pollutant emissions. Unfortunately, the lockdown-driven global ambient benzene changes still remained unknown. The ensemble machine-learning model coupled with the chemical transport models (CTMs) was applied to estimate global high-resolution ambient benzene levels. Afterwards, the XGBoost algorithm was employed to decouple the contributions of meteorology and emission reduction to ambient benzene. The change ratio (P<sub>dew</sub>) of deweathered benzene concentration from pre-lockdown to lockdown period was in the order of India (−23.6 %) > Europe (−21.9 %) > United States (−16.2 %) > China (−15.6 %). The detrended change (P*) of deweathered benzene level (change ratio in 2020 – change ratio in 2019) followed the order of India (P* = −16.2 %) > Europe (P* = −13.9 %) > China (P* = −13.3 %) > United States (P* = −6.00 %). Substantial decreases of atmospheric benzene levels saved sufficient health benefits. The global average lifetime carcinogenic risks (LCR) and hazard index (HI) decreased from 4.89 × 10−7 and 5.90 × 10−3 and 4.51 × 10−7 and 5.40 × 10−3, respectively.</p>