<p>Lung cancer remains the leading cause of cancer-related mortality worldwide, highlighting an urgent need for improved therapeutic strategies and more physiologically relevant in vitro models to better understand lung tumor biology and accelerate drug screening. Traditional in vivo tumor-bearing grafting model, despite their biological complexity, present significant limitations due to interspecies differences that hinder accurate recapitulation of the human tumor microenvironment (TME) and drug responses. On the other hand, existing in vitro models often fail to fully replicate the dynamic biochemical and biomechanical cues that are present in vivo. However, more recently, three-dimensional (3D) in vitro models, including organoids, tumoroids, and spheroids, have emerged as promising models that mimic TME. The conventional tumoroid generation relies on costly commercial extracellular matrix (ECM). Here, we introduced a microfluidic chip-based lung tumoroid model capable of mimicking alveolar stretching and air–liquid interface (ALI). Our model integrates three types of cells: lung cancer, endothelial, and fibroblast cells, enabling their self-aggregation into tumoroids without extragenous ECM. We also compared tumoroid formation capacity among three lung cancer cell lines, revealing that NCI-H1299, highly metastatic cells, failed to form tumoroids due to low E-cadherin expression. Furthermore, the drug testing demonstrated that the tumoroids exposed to ALI and cyclic stretch exhibited increased drug resistance compared to the static control, emphasizing the importance of mechanical cues in modulating tumor behaviors. This new in vitro model offers a more physiologically relevant platform to study lung cancer biology and drug screening, potentially accelerating the development of effective lung cancer therapies.</p>