One executive is touting open-weight models, especially if your company produces lots of data.
In an interview with CNBC Squawkbox, released on Tuesday, Hims & Hers CEO Andrew Dudum said companies that already have large datasets should shift away from using big AI models and toward using open-weight models.
He cited his company, a telehealth provider that delivers prescription drugs and personal care products through a subscription-based service. He said the company has a large closed-loop dataset of patients it has worked with, which he called a “real asset” because it can be used to train AI models.
“I think for companies that have the resources and scale, if they have an independent dataset, that is a path that they will go, no question,” he said to CNBC host Andrew Ross Sorkin. “The cost can be upward of 70, 80% less.”
Open-weight models let users access and customize a model’s trained parameters, and can be cheaper to run at scale. More importantly, Dudum said models trained on a company’s actual use cases would perform better.
“And so what we launched, our first version, immediately outsurpassed what we could get in market,” he said. “But the trajectory of what’s possible in just six months is transformative because it’s continuously learning on every new patient that comes into Hims & Hers.”
Dudum’s comments come as saving AI costs is one of the hottest topics in the tech and business spheres. Companies are phasing out tokenmaxxing, in which employees were urged to burn as many AI tokens as possible, and are now focusing on getting the most value from their spending.
Many are model routing, referring to matching tasks to AI models based on their complexity. Some executives, like Coinbase’s Brian Armstrong, have implemented the use of Chinese models like Kimi K2.7 and GLM 5.2 as default models within their companies.
Kimi K3, an open-weight model developed by the Chinese AI lab Moonshot AI, made waves in Silicon Valley last month for promising to rival the capabilities of models from OpenAI and Anthropic at lower costs.
Meanwhile, data has become one of the most sought-after resources in the AI space, as it is crucial to model training. Earlier this month, Google paid $10 million for internal data from the shuttered carrier Spirit Airlines, which included internal documents, workflows, emails, and millions of Teams messages.
AI training firm Handshake AI has promised to pay $6 per page of “high-quality” work documents, capped at $30,000, in a bid to gain access to more data.
