StudentBench: AI Tutors Match Human GRE Gains at 918x Lower Cost

AI tutoring produced GRE learning gains statistically equivalent to those from expert human tutors, at a cost the authors put at 918 times lower per percentage point of score improvement, according to StudentBench, a new arXiv paper and public platform.

The study ran 2,383 participants through AI tutoring, human tutoring, or no tutoring, and logged more than 175,000 student-AI messages. "AI tutoring is statistically equivalent to expert human tutoring for GRE learning gains (p = .015)," write the authors — Curtis Northcutt, Inaara Hasmani, Kevin Feng, Trevor Khangi, Andreas Plesner and Jonas Mueller — adding that "in five of the seven GRE domains, the best performing AI tutor surpassed the human tutor, on average."

The cost gap is where operator interest lands. The paper reports one AI tutor hit parity "at 918 times lower cost (USD 0.0052 for AI versus USD 4.81 for human, per percentage point gained)." For Quantitative sessions, the abstract flags a chain of correlations: "faster AI replies correlated with more student messages, more messages with more correct practice, and more correct practice with larger learning gains (all p < .002)."

A second study had expert tutors run 2,028 pairwise rubric comparisons of LLM-generated lesson plans and practice problems, separating the AI tutors across lesson planning, practice-problem creation, conversational pedagogy, cost and engagement. The abstract names no specific LLMs and publishes no per-model breakdown; the platform is freely available at studentbench.org. It is the latest entry in a thin but growing education beat we track, after OpenAI's Australian teen-AI blueprint last weekend.