This paper introduces Vision-Language Models (VLMs), which combine visual and linguistic understanding. VLMs have shown significant potential in various applications, from visual assistants to generative models. However, there are challenges in developing reliable VLMs, such as understanding spatial relationships and attributes, and avoiding hallucinations. The paper discusses different families of VLMs, including contrastive-based, masking-based, and generative-based models. It also covers training methods, data curation, and evaluation techniques for VLMs. The paper highlights the importance of data quality, training strategies, and evaluation benchmarks in improving VLM performance. It also discusses the extension of VLMs to video and the challenges involved. The paper concludes by emphasizing the need for responsible development of VLMs and the importance of understanding the underlying mechanisms of these models.This paper introduces Vision-Language Models (VLMs), which combine visual and linguistic understanding. VLMs have shown significant potential in various applications, from visual assistants to generative models. However, there are challenges in developing reliable VLMs, such as understanding spatial relationships and attributes, and avoiding hallucinations. The paper discusses different families of VLMs, including contrastive-based, masking-based, and generative-based models. It also covers training methods, data curation, and evaluation techniques for VLMs. The paper highlights the importance of data quality, training strategies, and evaluation benchmarks in improving VLM performance. It also discusses the extension of VLMs to video and the challenges involved. The paper concludes by emphasizing the need for responsible development of VLMs and the importance of understanding the underlying mechanisms of these models.