Pinterest is recalibrating its approach to artificial intelligence by reducing its AI budget even as it continues to expand its AI-supported tools for users and advertisers. This strategic shift comes amid rising pressures on companies to demonstrate tangible returns on their investments in AI technologies.
Vicky Gkiza, the vice president of product management at Pinterest, outlined the company’s “model-agnostic” strategy, initiated in 2023. This approach blends Pinterest’s own AI models with closed-source models from companies like Anthropic and OpenAI, as well as open-source options from Alibaba. This combination aims to leverage the strengths of each model type while managing investment costs.
Closed-source models are known for their ability to quickly process significant data volumes and seamless integration into existing systems, albeit at a higher cost. In contrast, open-source models are free to use and can be tailored extensively, although they necessitate skilled software engineers for maintenance and optimization. Lan Guan, chief AI and data officer at Accenture, noted that firms are increasingly recognizing the benefits of a multimodal AI approach, which strikes a balance between operational performance and cost management associated with token usage in AI processing.
As part of its strategy, Pinterest has engaged in beta testing while also updating its hiring practices to emphasize software development skills. These efforts have been pivotal in enhancing its generative AI offerings, according to Gkiza. Pinterest uses OpenAI’s closed-source large language models to support various product features while employing Anthropic’s Claude for internal needs like coding. Furthermore, Alibaba’s Qwen, an open-source large language model, assists in visual and content understanding.
By mid-2025, Pinterest successfully launched several AI features that incorporated this blended AI approach. For example, the auto-collages feature, which allows advertisers to transform their product catalogs into pins for shopping feeds, began testing in early 2024. By June 2025, it was ready for a pilot program that included retailer-advertisers such as Macy’s. Additionally, Pinterest’s voice-enabled AI feature, which utilizes both open-source and third-party AI, was beta tested in October 2025. Early feedback indicated that users were more inclined to ask shopping-related queries verbally rather than typing, showcasing the evolution of search dynamics on the platform.
The company is not only focusing on technology but also on talent acquisition. Gkiza emphasized the importance of hiring experts in AI and machine learning to enhance the customization of its large language models. Notably, former Google engineer Matthias Zenger was appointed vice president of engineering in April 2025, followed by the recruitment of software engineer Mirjam Wattenhofer to enhance user experience in e-commerce. Both are based out of a new Engineering Excellence Hub in Zurich, which focuses on refining user experiences utilizing AI and machine learning.
In its February 2026 earnings presentation, Pinterest highlighted that its blended AI methodology costs approximately 90% less than when it relied solely on proprietary models. The company anticipates further investments in cloud-computing infrastructure, including graphics processing units, which are essential for the effective operation of its cost-saving technology.
Looking ahead, Pinterest plans to continue experimenting with various AI models, prioritizing its proprietary models for personalization, open-source models for cost efficiency, and closed-source models for performance optimization. This strategic focus not only illustrates Pinterest’s commitment to enhancing user experience but also positions it to adapt to the rapidly changing landscape of artificial intelligence.
See also
OpenAI’s GPT Image2 Generates MS Paint-Style Drawings, Sparking Viral Trend
Apple Researchers Reveal LaDiR Framework, Enhancing LLM Accuracy by 20% in Math and Code Generation
Bangalore AI Startups Develop Proprietary Models, Target Global Markets with Local Solutions
AI Threatens Electoral Integrity in Nigeria’s 2027 Elections Amid Rising Misinformation Risks
OpenAI Tests GPT 5.6 in Codex Update to Enhance AI Coding and Cybersecurity Features
