In pursuing the development of AI-powered tools for users and advertisers, Pinterest leaders said they are actually scaling back some of their AI budget.
As companies continue to invest in AI, they face pressure from investors to demonstrate return on investment from this rapidly evolving technology. One way to do this is to reduce the budget and increase investment needed in systems that enhance AI capabilities.
Pinterest, a visual-first social media platform where users can save images and curate ideas, is taking a so-called “model agnostic” approach to generative AI, said Vicky Gkiza, the company’s vice president of product management. The strategy will begin in 2023 and includes combining Pinterest’s own AI models developed by the company’s software engineers, Anthropic and OpenAI’s closed source models, and Alibaba’s open source models, Gkiza said.
Closed-source AI models process large amounts of data quickly, require little maintenance, and their pre-built nature allows for immediate integration into a company’s existing systems, but they are typically more expensive. On the other hand, open source models are typically free to download, use publicly available data, and can be modified to a greater extent than closed source products. However, it also requires the expertise of software engineers who can build, maintain, and debug these highly customizable models.
Lan Guan, Accenture’s chief AI and data officer, said that enterprises are increasingly seeing value in multimodal AI approaches that balance performance and token cost, or process units of text AI models. “If we don’t start actively managing our tokens, the cost of this token will slow down,” Guan told Business Insider. “Open source would be a really good option.”
Gkiza spoke to Business Insider about how Pinterest has strengthened its blended approach to generative AI through beta testing, a modern hiring strategy focused on software development expertise, and investments in infrastructure.
technology
Pinterest uses OpenAI’s closed-source large-scale language models to support some of its product features, and relies on another closed-source AI, Anthropic’s Claude, for internal use cases such as coding, a company spokesperson told Business Insider. Alibaba’s Qwen, an open-source LLM, is used for visual and content understanding, data labeling, and assistant tasks.
With this approach, Gkiza said the company rolled out several new AI capabilities in 2025, each blending open-source and closed-source generative AI.
Auto collage. A feature that allows advertisers to turn their product catalogs into pins that appear in shopping feeds in categories like skin care, home decor, fashion, and more. Pinterest began testing Autocollage in early 2024, with software engineers fine-tuning it using a combination of in-house development, open source, and third-party AI models. By June 2025, Autocollage was ready for pilots that included a small group of retailer-cum-advertisers like Macy’s.
Pinterest’s voice-enabled AI feature, which uses both open source and third-party AI to generate responses to user queries, then underwent beta testing in October 2025. Early results show that users tend to ask more shopping-specific questions when they can ask them out loud rather than typing a search, a Pinterest spokesperson said.
“Search is evolving very quickly, and using AI to improve it was essential,” Gkiza told Business Insider.
talent
To evolve new AI-enabled features and implement a blended model approach, Pinterest also hired employees with AI and machine learning expertise to manage language model customization at scale, Gkiza said. Matthias Zenger, a former Google engineer, joined the company in April 2025 as vice president of engineering. Three months later, in August, Pinterest announced it was hiring software engineer Mirjam Wattenhofer and opening an Engineering Excellence Hub in Zurich to focus on e-commerce and user experience.
Zenger and Wattenhofer both work out of the Zurich office, where the team focuses on using AI and machine learning technology to improve the user experience. Pinterest CEO Bill Ready said in February that the company would hire additional research and development personnel to support its AI efforts.
“We’re investing more in hiring the right people and evolving our teams, whether it’s engineering or product management, to become more AI-savvy,” Gkiza said.
result
Gkiza said the company’s blended approach to AI-driven experiences is costing an estimated 90% less than when Pinterest relied solely on its proprietary model, a milestone the company touted in its fourth quarter earnings call in February 2026.
With the increased use of open source AI models, the company plans to invest more in cloud computing infrastructure, such as graphics processing units, needed to power its cost-saving technology, Gkiza said.
Pinterest said it will continue to experiment with different AI models, prioritizing in-house models for personalization, open-source models for cost-effectiveness and multimodal machine learning, and closed-source models where they still provide the best performance.