Interview
Open-source AI vs. frontier models: DataCamp’s real-world test

Jonathan Cornelissen
Co-Founder & CEO, DataCamp

Yusuf Saber
Chief AI Officer, DataCamp
51 minHosted by Kim Isenberg and Peter Thum
Watch on YouTubeAbout this conversation
Open models can be dramatically cheaper than frontier APIs — but the price per token is only one part of the real cost of running AI in production. Kim Isenberg and Peter Thum talk with Jonathan Cornelissen, Co-Founder and CEO of DataCamp, and Yusuf Saber, DataCamp’s Chief AI Officer, about the technology behind DataCamp’s AI-native tutor on a learning platform serving more than 19 million people.
Rather than discussing open-source AI in theory, the conversation focuses on what happens when real users, infrastructure costs, teaching quality, latency, privacy and vendor lock-in enter the equation.
Topics
- How DataCamp’s AI tutor generates and personalizes lessons
- How tasks are routed between open-weight and frontier models
- Whether open models are really around 10x cheaper
- The hidden engineering and infrastructure costs behind self-hosting
- How an AI tutor avoids confidently teaching incorrect information
- Privacy, data residency and dependence on model providers
- What companies should measure before choosing their AI stack
Kim Isenberg is co-founder and Editor-in-Chief of Superintelligence, an AI newsletter published every weekday.
Read the NewsletterMore interviews
- Xianming LiuXPeng’s AI chief on VLA 2.0, robotaxis and physical AI
- Philipp Türker & Marco SzeidenlederEurope doesn’t want AI sovereignty — it wants US AI with a local postcode
- Stephanie CohenCan Cloudflare make AI companies pay creators?