Inside Pinterest's efforts to replace expensive AI with open-source models

• Pinterest's VP of product management said mixing open-source and closed-source AI models helped cut costs.
• The social platform's recent AI-enabled features combine OpenAI, Alibaba, and its own AI models.
• To support this approach, Pinterest invested in cloud infrastructure and an updated hiring strategy.

In its pursuit of creating AI-supported tools for its users and advertisers, Pinterest leaders said it is actually shrinking part of its AI budget.

As companies continue to spend on AI, they face pressure from investors to show returns on investment from this fast-evolving technology. One way to do that: Stretch investments by shrinking the budget needed for the systems that power AI features.

Pinterest, the visual-first social media platform where users can save images to curate their ideas, is taking what it calls a "model-agnostic" approach to generative AI, said Vicky Gkiza, the company's vice president of product management. The strategy started in 2023 and involves combining Pinterest's proprietary AI models, developed by the company's software engineers, its closed-source models from Anthropic and OpenAI, and open-source models from Alibaba, said Gkiza.

While closed-source AI models can quickly process large amounts of data, require little maintenance, and integrate instantly into a company's existing systems due to their pre-built nature, they're typically more expensive. Open-source models, meanwhile, are typically free to download, use publicly available data, and can be modified more deeply than closed-source offerings. However, they also require the expertise of software engineers who can build, maintain, and debug these highly customizable models.

Lan Guan, the chief AI and data officer at Accenture, said companies are increasingly seeing value in a multimodal AI approach that balances performance and the cost of tokens, or the units of text AI models process. "This token cost is going to slow you down if you don't start managing them proactively," Guan told Business Insider. "Open-source will be a really good option."

Gkiza walked Business Insider through how Pinterest bolstered its blended approach to generative AI through beta testing, an updated hiring strategy focused on software development expertise, and infrastructure investments.

The tech

Pinterest uses OpenAI's closed-source large language models to support some of its product features and relies on Anthropic's Claude, another closed-source AI, for internal use cases like 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, the company rolled out several new AI features in 2025, each blending open-source and closed-source generative AI, said Gkiza.

Auto-collages, a feature where advertisers can convert their product catalogs into pins that populate on shopping feeds for categories like skincare, home decor, and fashion. Pinterest began testing auto-collages at the beginning of 2024, when software engineers fine-tuned a mix of internally developed, open-source, and third-party AI models. By June 2025, auto-collages was ready for a pilot, which included a small group of retailer-advertisers like Macy's.

Then, Pinterest's voice-enabled AI feature, which uses both open-source and third-party AI to generate responses to users' queries, underwent beta testing in October 2025. Early results showed that users tended to pose more shopping-specific questions when they could ask aloud rather than typing their searches, said a Pinterest spokesperson.

"Search has been evolving so fast. It was imperative for us to use AI to improve," Gkiza told Business Insider.

The talent

To advance its new AI-enabled features and execute its blended-model approach, Pinterest also hired employees with AI and machine learning expertise to steward the customization of its large language models, said Gkiza. Former Google engineer Matthias Zenger joined the company as its vice president of engineering in April 2025. Three months later, in August, Pinterest announced it had hired software engineer Mirjam Wattenhofer to focus on e-commerce and user experiences, and that it would open an Engineering Excellence Hub in Zurich.

Both Zenger and Wattenhofer work at the Zurich office, where a team focuses on improving user experiences with AI and machine-learning technologies. In February, Pinterest CEO Bill Ready said the company would hire additional research and development workers to support the company's AI efforts.

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

The outcome

Gkiza said the company's blended approach to AI-driven experiences costs an estimated 90% less than when Pinterest relied solely on its proprietary models, a milestone the company touted during its February 2026 fourth-quarter earnings presentation.

With its growing use of open-source AI models, the company plans to invest more in cloud-computing infrastructure, like graphics processing units, which are needed to power the cost-saving technology, Gkiza said.

Pinterest said it will continue to experiment with various AI models, prioritizing its own models for personalization, open-source models for cost-effectiveness and multimodal machine learning, and closed-source models when they still offer the best performance.