The news: Pinterest was able to drive down its AI costs 90% while improving its model’s accuracy by 30% by removing Alibaba’s Qwen3-VL’s vision layer and rebuilding it with proprietary data, per Venture Beat.
Pinterest has been investing heavily in customizing open-source models “foundationally in-house.” Fine-tuning open models for its own use cases lets Pinterest pull metadata from pins and images, run computations offline, and retrain the model as needed.
“If you’ve got really unique data that you can then fine-tune an open-source model with, data quality will, frankly, outweigh or overcome model size,” Pinterest CTO Matt Madrigal said in a recent Venture Beat podcast.
Why it’s worth watching: AI inference and token costs are upending established marketing pricing models, per Digiday, but Pinterest’s example reveals that in-house teams can dramatically drive down costs and improve accuracy.
The caveat: This strategy only works on open-source large language models (LLMs) like Qwen, Meta’s Llama, or DeepSeek.
Implications for brands: Brands should look into open models for cost-efficiency and customization and also keep in mind that security and reliability are no longer the AI provider’s responsibility once the LLM has been altered.
Improved model accuracy, deeper customization, and reduced costs could be compelling reasons for brands to start experimenting and offloading AI workflows to free up time and capital for building in-house teams and solutions.
This content is part of EMARKETER’s subscription Briefings, where we pair daily updates with data and analysis from forecasts and research reports. Our Briefings prepare you to start your day informed, to provide critical insights in an important meeting, and to understand the context of what’s happening in your industry. Non-clients can click here to get a demo of our full platform and coverage.
You've read 0 of 2 free articles this month.
685 Third Avenue21st FloorNew York, NY 100171-800-405-0844
1-800-405-0844[email protected]