Pinterest CEO touts open source AI: ‘tremendous performance’ with reduced costs (techcrunch.com)

Pinterest CEO Bill Ready told investors that the company is seeing “tremendous performance” from fine‑tuned open‑source models for its visual AI use cases, delivering comparable results to proprietary models at “orders of magnitude” lower cost. The announcement came on a Q3 earnings call that also revealed softer holiday revenue guidance and a steep stock drop, but Ready emphasized that Pinterest regularly benchmarks off‑the‑shelf proprietary models against open‑source alternatives and plans to adopt many open‑source models to support features like multimodal search (text+image), personalized recommendations, ad targeting, AI product discovery, the Pinterest Assistant, and AI‑curated boards. For the AI/ML community this is a concrete production example that open‑source models, when properly fine‑tuned for domain and modality (visual + textual signals), can materially reduce inference spend while preserving performance—shifting the economics of deploying large models at scale. It underscores the importance of domain adaptation, multimodal architectures, and cost‑aware inference strategy for consumer platforms, and signals continued caution around fully agentic commerce: Pinterest will evaluate users’ willingness to let AI “push the button” even as it refines assistant-driven discovery and monetization pathways.