AI Tutor Matches Human Expert on GRE Score Gains

In a randomized study by Handshake AI with 2,383 students, learning with an LLM tutor produced a statistically equivalent GRE score gain compared to a human expert, while Gemma 4 31B achieved the same result 918 times cheaper.

Compiled by Sergey KostenchukPublished 2026-09-25Updated 2026-09-25

2026-09-25 Research HuggingFace

🤖 AI Tutor Matches Human Expert on GRE Score Gains

Handshake AI released StudentBench (arXiv:2609.28470): 2,383 students, 2,139 AI sessions versus 140 with a human expert. The GRE gain from an LLM tutor is statistically equivalent to lessons with a $75/hour tutor (p = .015), with AI outperforming humans in 5 of 7 domains. Gemma 4 31B delivers the same result 918 times cheaper: $0.0052 versus $4.81 per point.

🌍 StudentBench is the first to demonstrate the economic equivalence of an AI tutor to a human expert and shifts the focus from 'what the model knows' to 'what it can teach' — an argument against the 'illusion of learning' and in favor of mass AI tutoring.

👤 Code, data (176,614 messages), and reproduce.py are open on GitHub and Hugging Face; the platform is studentbench.org. The AI tutor already competes with humans on feedback speed and is unmatched on price.

Source 1: https://arxiv.org/abs/2609.28470 Source 2: https://github.com/Handshake-AI-Research/studentbench

Sources