StudentBench: AI and human tutoring yield equivalent GRE learning gains

TL;DR

StudentBench is a public platform enabling large-scale evaluation of AI tutoring through over 175,000 student-AI interactions. It demonstrates that AI tutoring produces learning gains on the GRE equivalent to human tutoring, with some AI tutors outperforming humans in specific domains and achieving comparable results at drastically lower cost.

Why It Matters

This research provides empirical evidence that AI tutors can match human tutors in educational outcomes, offering a scalable, cost-effective alternative for high-stakes test preparation. The findings support the integration of AI into education systems where access to expert human tutors is limited.

Key Findings

• AI tutoring produced statistically equivalent learning gains to human tutoring on the GRE, with p-values of .015 and .044 in two studies.
• One AI tutor achieved equivalent learning outcomes at 918 times lower cost per percentage point gained compared to human tutoring.
• Faster AI response times correlated with increased student engagement, more correct practice responses, and higher learning gains in Quantitative GRE sessions.