The paper argues that tabular foundation models (LTMs) should be a research priority, as they are underexplored despite their widespread use in many fields. Unlike text and image foundation models, which have received significant attention, tabular data is often overlooked due to its complexity, lack of structured meaning, and challenges in handling mixed data types. LTMs could revolutionize how tabular data is used, enabling contextual analysis across related datasets and offering applications in data science, synthetic data generation, and multidisciplinary scientific discovery. The paper highlights the potential of LTMs to address real-world challenges, including data scarcity, bias, and privacy concerns, while also improving the efficiency and accuracy of data analysis. LTMs are defined as large-scale models capable of handling tabular data, with requirements including mixed-type columns, cross-dataset modeling, textual context, and invariance to column order. Current research on LTMs is limited, with most focusing on representation learning and supervised learning, while generative models remain underdeveloped. The paper discusses the challenges of building generative LTMs, including the difficulty of modeling continuous variables and the need for efficient, scalable architectures. The paper also emphasizes the real-world impact of LTMs, including their potential to improve inclusivity, data privacy, and reproducibility. LTMs could help address data scarcity in underrepresented groups, generate synthetic data for bias mitigation, and support scientific research through meta-analyses and data integration. Additionally, LTMs could serve as a powerful tool for data scientists, enabling automated data analysis, cleaning, and visualization. The paper concludes that while LTMs are still in their early stages, they offer significant potential for impact across various domains. However, challenges remain in terms of data quality, model evaluation, and bias mitigation. The paper calls for increased research into LTMs, emphasizing their importance in advancing the field of machine learning and data science.The paper argues that tabular foundation models (LTMs) should be a research priority, as they are underexplored despite their widespread use in many fields. Unlike text and image foundation models, which have received significant attention, tabular data is often overlooked due to its complexity, lack of structured meaning, and challenges in handling mixed data types. LTMs could revolutionize how tabular data is used, enabling contextual analysis across related datasets and offering applications in data science, synthetic data generation, and multidisciplinary scientific discovery. The paper highlights the potential of LTMs to address real-world challenges, including data scarcity, bias, and privacy concerns, while also improving the efficiency and accuracy of data analysis. LTMs are defined as large-scale models capable of handling tabular data, with requirements including mixed-type columns, cross-dataset modeling, textual context, and invariance to column order. Current research on LTMs is limited, with most focusing on representation learning and supervised learning, while generative models remain underdeveloped. The paper discusses the challenges of building generative LTMs, including the difficulty of modeling continuous variables and the need for efficient, scalable architectures. The paper also emphasizes the real-world impact of LTMs, including their potential to improve inclusivity, data privacy, and reproducibility. LTMs could help address data scarcity in underrepresented groups, generate synthetic data for bias mitigation, and support scientific research through meta-analyses and data integration. Additionally, LTMs could serve as a powerful tool for data scientists, enabling automated data analysis, cleaning, and visualization. The paper concludes that while LTMs are still in their early stages, they offer significant potential for impact across various domains. However, challenges remain in terms of data quality, model evaluation, and bias mitigation. The paper calls for increased research into LTMs, emphasizing their importance in advancing the field of machine learning and data science.