At a time when AI costs are soaring and large model calls can run into millions, Pinterest is quietly forging a "cost-effective intelligent path." CEO Bill Ready revealed in the latest earnings call that by extensively adopting fine-tuned open-source large models, Pinterest has achieved performance comparable to closed-source top models in visual AI tasks, while reducing costs by an order of magnitude. This strategy not only effectively mitigated the pressure from slowing ad revenue growth, but may also provide a new cost-reduction and efficiency-enhancement model for the entire e-commerce and content recommendation industry.

As the "inspiration starting point" of users' shopping journey, Pinterest has long relied on AI-driven personalized recommendations, mixed-text-and-image search, ad targeting, and new product discovery. The newly launched Pinterest Assistant is a key product in its transformation into an "AI shopping partner"—users can directly interact with AI to receive suggestions for fashion, home decor, or gift items based on their personal boards, browsing habits, and preferences of similar users.

However, when faced with investor questions about the commercial prospects of "agentic commerce," Ready remained cautious and pragmatic. He stated that although the platform has already enabled "one-click purchase" through collaboration with Amazon, whether to let AI automatically place orders for users still needs to observe user real intentions. "We focus more on guiding users to make decisions, rather than acting on their behalf," he emphasized.

Open-source models become a cost-cutting tool, closed-source APIs are marginalized

Ready specifically pointed out that Pinterest continuously compares mainstream closed-source models with open-source solutions internally, and the results were encouraging: "In our visual AI scenarios, fine-tuned open-source models performed extremely well. Considering the current market price per token, their cost is just a fraction of closed-source solutions, yet the performance is no less impressive." Based on this, the company has decided to fully shift to deploying open-source models in multiple core scenarios, significantly optimizing the long-term cost structure.

This decision comes at an opportune time. Due to the impact of the new tariff policy from the Trump administration, Pinterest expects Q4 revenue to be slightly below market expectations (13.1–13.4 billion USD vs 13.4 billion expected), and the stock plummeted over 21% in a single day. Under this pressure, ROI (return on investment) of AI spending has become particularly critical. The low cost and high controllability of open-source models have provided Pinterest with a technical lever to "maintain profitability without sacrificing the user experience."

In addition, the platform is advancing an AI plus human collaborative personalized dashboard, where algorithms initially filter content, then hand it over to domain experts for refinement, balancing scale efficiency and aesthetic quality. Ready said that this hybrid model can enhance user engagement and create higher value display scenarios for brand advertisers.

AIbase believes that Pinterest's strategy sends a strong signal: while the industry is still anxious about the high bills of closed-source APIs, the combination of open-source models, scenario-specific fine-tuning, and edge-side optimization has become the preferred approach for practical enterprises. In the deep waters where AI is moving from "showcasing" to "monetization," the ability to control costs and focus on vertical scenarios may be more decisive for a company's survival and growth than the size of parameters.