Switch to Dark Mode
Carnegie Learning's MATHia — an AI tutoring platform born out of Carnegie Mellon University — doubled student math learning in a 140-school randomised trial and now serves 5.5 million students across all 50 US states. Here is how it works and what the evidence says.
July 3, 2026
Info
Math has a long-standing equity problem in schools. A one-size-fits-all classroom pace means that students who fall behind rarely catch up, while students who are ahead disengage. One-to-one tutoring is the most effective known intervention, but it is expensive, hard to staff at scale, and unavailable to most students. The achievement gap it could close has persisted precisely because the solution has never been scalable.
Carnegie Learning was spun out of Carnegie Mellon University's Human-Computer Interaction Institute, where researchers built the first AI-based cognitive tutor in the 1990s. The core idea: instead of presenting content and testing recall, model what the student actually knows and adapt to it in real time. MATHia, the company's flagship platform for grades 6-12, is the commercial evolution of that research. It does not serve pre-written problem sets on a fixed path. It builds a live model of each student's understanding, identifies precisely where their knowledge breaks down, and generates instruction targeted at that specific gap.
Unlike most edtech, Carnegie Learning has submitted its platform to rigorous independent evaluation over decades. The landmark result is a randomised controlled trial across 140 schools, which found that students using the AI Cognitive Tutor course doubled their math learning compared to students following their regular curriculum. A 2025 study published in Scientific Reports found that AI tutoring outperformed in-class active learning with an effect size of 0.73 to 1.3 standard deviations — a large effect by any measure in education research. Broader research on AI tutoring programmes at this scale also reports a 70% improvement in course completion rates and a 15% reduction in dropout.
Most education technology automates delivery — it gets content in front of students faster and more conveniently. MATHia automates the cognitive model. The distinction matters enormously. A video lesson or a digital quiz presents the same experience to every student. A cognitive tutor that actually models each student's knowledge state can do something a teacher with 30 students in the room cannot: give every student, simultaneously, the specific next problem they are ready for and the precise hint they need when they get stuck. That is the mechanism behind the learning gains. The AI is not replacing the teacher. It is doing the individualised diagnostic work that no teacher has the bandwidth to do at scale, and surfacing the results so teachers can direct their attention where it is needed most.
The AI is not replacing the teacher. It is doing the individualised diagnostic work that no teacher has the bandwidth to do at scale — and surfacing the results so teachers can direct their attention where it is needed most.