StudentBench - Learning Gain: leaderboard

Real students take a GRE practice pre-test, are tutored by an AI model for an hour, then take a different post-test; the score is the pretest-adjusted learning gain in percentage points for students of that tutor. Each row is one model at one reasoning setting. Higher is better. Source: studentbench.org.

Metric: GRE learning gain after one hour of AI tutoring (percentage points, pretest-adjusted ANCOVA). Source: studentbench.org. Saturation forecast: Around January 2028. 11 models tracked. Updated 4 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 4.8xHigh16#64
2GPT-5.5 ProMedium15.7#40
3Gemini 3.1 Pro (Preview)High15.4#101
4Gemini 3.5 FlashLow14.3#108
5Gemini 3.7 FlashMedium14.1#42
6GPT-5.5High13.5#54
7Gemma 4 31BHigh13.1#389
8GPT-5.4 MiniNon-reasoning13#405
9Claude Opus 5High12.6#26
10Kimi K2.6Medium12.2#192
11Gemini 3.6 FlashLow12.1#79

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=studentbench-learning-gain · How It Works · Data refreshed daily, snapshot 2026-11-20.