Pinterest cuts AI costs 90% by mixing open-source and proprietary models

Pinterest cuts AI costs 90% by mixing open-source and proprietary models

Pinterest combined proprietary, open-source, and closed-source AI models to reduce the cost of its AI-enabled features by an estimated 90%, the company announced during its fourth-quarter earnings presentation in February 2026.

The social platform's approach reflects a broader shift among companies struggling to justify AI spending to investors. Rather than relying on expensive closed-source models alone, Pinterest built a strategy that balances cost, performance, and customization needs.

How the model mix works

Vicky Gkiza, Pinterest's vice president of product management, said the company uses OpenAI's models for some product features and Anthropic's Claude for internal coding tasks. For cost-sensitive work like visual understanding and data labeling, Pinterest relies on Alibaba's Qwen, an open-source language model.

Closed-source models process large amounts of data quickly and integrate easily into existing systems. They cost significantly more. Open-source models are free to download and can be modified deeply, but require software engineers to build, maintain, and debug them.

Lan Guan, chief AI and data officer at Accenture, said companies increasingly see value in balancing performance against token costs-the units of text that AI models process. "Open-source will be a really good option," Guan said.

Features in production

Pinterest rolled out auto-collages in mid-2025, a feature that converts product catalogs into pins for shopping feeds. The company tested the feature starting in early 2024 with a mix of internal, open-source, and third-party models before piloting with retailers like Macy's.

A voice-enabled search feature launched in beta in October 2025. Early results showed users asked more shopping-specific questions when speaking aloud rather than typing.

Hiring and infrastructure

Pinterest hired engineers with machine learning expertise to customize its language models. Former Google engineer Matthias Zenger joined as vice president of engineering in April 2025. Three months later, the company hired software engineer Mirjam Wattenhofer to focus on e-commerce experiences and opened an Engineering Excellence Hub in Zurich.

Gkiza said the company is "investing more in hiring the right talent - evolving the team, whether it is engineering or product management - to be much more familiar with AI."

Pinterest plans to invest in cloud infrastructure, including graphics processing units, to support its open-source models. The company prioritizes its own models for personalization, open-source models for cost-effectiveness, and closed-source models when they deliver the best performance.

Related: Learn more about Generative AI and LLM strategies and AI for Product Development.