Pinterest, a major player in visual discovery, has shifted its strategy to focus on open-source AI models to enhance efficiency and reduce costs. This change comes after the company recognized the evolving capabilities of open-source technology, which now offers performance comparable to that of leading frontier AI models at a fraction of the price—less than 10% of the cost. This strategic pivot has led to the integration of models that it has developed in-house for specific applications, enabling more tailored solutions for their unique needs. The decision to pivot towards open-source was driven predominantly by economic factors. Pinterest aimed to maintain user experience and service velocity while navigating the challenges posed by high costs associated with frontier models. The company acknowledges that they still utilize some frontier models, including OpenAI, for certain internal processes and features. However, the emphasis on open-source allows for fine-tuning capabilities that expedite their development cycle while ensuring they remain user-focused. One notable application of their open-source efforts is the model named Queen, which assists in analyzing visual content and generating descriptive text when metadata is lacking. This capability is essential for improving user interaction on the platform. Moreover, with the launch of Pinterest Assistant—a multimodal chat feature—Pinterest further illustrates how these open-source models can enhance their services by providing personalized shopping recommendations through various mediums such as images, voice, and text. Despite the advantages of employing open-source AI, there are inherent challenges that Pinterest must navigate, including the need for careful scrutiny of open-source code for potential vulnerabilities. The sustainability of open-source projects is also a concern, especially as AI-driven contributions increase the volume of low-quality input. Thus, Pinterest has expressed commitment to the ecosystem through ideas of contributing back to the open-source community by releasing research, participating in conferences, and possibly sharing their own models. These efforts are geared towards fostering a healthier open-source landscape while balancing its need for cost-efficient technology.