Hims & Hers Claims to Cut AI Costs with Open Models

Andrew Dudum, CEO of Hims & Hers, argues that open-weight models trained on proprietary data can significantly reduce AI costs. His statements come at a time when companies are seeking efficiency and monetizing access to their internal corpora.

Google Pays $10 Million for Data from Spirit Airlines

Google has spent $10 million on internal data from Spirit Airlines, including documents, workflows, emails, and millions of Teams messages. In the same spirit of the data race, Handshake AI has promised $6 per page for high-quality working documents, with a cap of $30,000. Data is emerging as a key resource for training models, while reducing AI costs is a hot topic in tech and business circles. Companies are moving away from tokenmaxxing to prioritize the best use of their spending and are adopting model routing, which aligns task complexity with suitable models.

Dudum Claims Up to 80% Savings with Open Weights

Andrew Dudum, CEO of Hims & Hers, recommends that organizations with vast data reserves abandon generalist AI models in favor of open-weight models. According to him, for large-scale companies with a distinct dataset, it is possible to reduce costs by 70 to 80%, potentially reaching 80%. These models, which allow for adjustable and consultable parameters, present a potentially less expensive option for large-scale deployment. Dudum also asserts that training models on real-world use cases leads to better performance. He cites Hims & Hers, a subscription telehealth provider, which reportedly has a closed-loop dataset on its patients and whose initial version immediately outperformed market offerings. He describes a "transformative" trajectory over six months, emphasizing continuous learning with each new patient.

Companies Adopt Model Routing and Chinese Models

Beyond optimizing usage, some leaders are defaulting to alternative models. Brian Armstrong at Coinbase has implemented the use of Chinese models such as Kimi K2.7 and GLM 5.2. Meanwhile, Kimi K3, developed by Moonshot AI, promises to compete with OpenAI and Anthropic models at a lower cost. Discussions around model choices and their costs are among the most debated topics in the tech ecosystem.