AI tutoring matches human tutors on GRE gains at 918 times lower cost

single source· 1 articles · confidence: high · first seen 2026-09-22 20:00 UTC

What this means for you

If you build or buy tutoring products, the metric this paper pushes is cost per point of test-score gain — $0.0052 in the AI condition against $4.81 for human tutors — rather than raw answer accuracy. Treat it as a preprint about GRE prep, not a settled result. Nothing to migrate.

A preprint on arXiv reports that AI tutoring produced learning gains statistically equivalent to expert human tutoring on GRE questions (a US graduate-admissions test), across 2,383 participants assigned to AI tutoring, human tutoring or none. The paper says the best AI tutor beat the human tutor on average in five of seven GRE domains. A second study ran 2,028 pairwise rubric evaluations, in which expert tutors graded machine-written lesson plans and practice problems against criteria. One AI tutor matched human gains at 918 times lower cost: $0.0052 against $4.81 per percentage point gained. The work covers GRE preparation only and is not peer reviewed.

Key facts

  • ·StudentBench measured GRE learning gains across 2,383 participants assigned to AI tutoring, human tutoring, or no tutoring. source
  • ·The paper reports AI tutoring statistically equivalent to expert human tutoring (p = .015), with the best AI tutor ahead on average in five of seven GRE domains. source
  • ·A second study collected 2,028 pairwise rubric evaluations from expert human tutors comparing LLM-generated lesson plans and practice problems. source
  • ·One AI tutor reached gains equivalent to human tutoring (p = .044) at 918 times lower cost: USD 0.0052 for AI versus USD 4.81 for human, per percentage point gained. source
  • ·The platform collected over 175,000 student-AI messages and is available free at studentbench.org. source

What the sources say

Sources

The original reporting. Follow these — they did the work.

← the wire