StudentBench: AI and human tutoring yield equivalent GRE learning gains

StudentBench finds AI tutoring statistically matched expert human GRE tutoring for 2,383 learners, with one AI tutor about 918 times cheaper per percentage point gained.

StudentBench is a public AI teaching-evaluation suite and data-collection platform built from more than 175,000 student-AI messages. Across 2,383 participants, AI tutoring was statistically equivalent to expert human tutoring on Quantitative and Verbal GRE questions (p = .015), and the best AI tutor surpassed the human tutor on average in five of seven domains. One AI tutor matched human learning gains (p = .044) at roughly 918 times lower cost, USD 0.0052 versus USD 4.81 per percentage point gained. In a second study, expert tutors compared LLM-generated lesson plans and practice problems in 2,028 pairwise rubric evaluations. Both the Hugging Face daily-papers item (2026-09-22) and the arXiv report (2026-09-23) report the same participant, significance, domain, and cost figures; the arXiv report is more specific about the GRE sections tested and describes the expert comparisons as rubric-based.

• 2,383 participants were measured on GRE learning gains using a platform with more than 175,000 student-AI messages.
• AI tutoring was statistically equivalent to expert human tutoring on Quantitative and Verbal GRE questions (p = .015).
• The best AI tutor outperformed the human tutor on average in five of seven GRE domains.
• One AI tutor matched human learning gains (p = .044) at roughly 918 times lower cost: USD 0.0052 versus USD 4.81 per percentage point gained.
• A second study collected 2,028 pairwise expert rubric evaluations of LLM lesson plans and practice problems.
• Sources: Hugging Face daily papers (2026-09-22) and arXiv cs.AI/cs.LG/cs.CL (2026-09-23); the two reports agree on all stated figures.