ATLAS-Education (StudentBench) Lesson Planning
Professional Work · 2026-09-23
Handshake AI's StudentBench (ATLAS-Education) lesson-planning leaderboard: 51 expert GRE tutors blindly compared pairs of AI tutors' lesson plans written for real students' pre-test results, rating concept relevance, grouping, prioritization, time allocation and test-taking strategies. A centered Bradley-Terry model fit to 9,577 non-tied ratings across 381 pre-tests gives each AI tutor configuration an ability in log-odds, so scores are relative to the 13 configurations compared.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 1.1 |
| GPT-5.5 Pro | 0.9 |
| Opus 4.8 | 0.9 |
| GPT-5.5 | 0.9 |
| Sonnet 4.6 | 0.4 |
| Sonnet 5 | 0.0 |
| Kimi K2.6 | -0.0 |
| Gemini 3.7 Flash | -0.4 |
| Gemini 3.1 Pro Preview | -0.9 |
| GPT-5.4 Mini | -1.0 |
| Gemma 4 31B IT | -1.1 |
| Gemini 3.5 Flash | -1.3 |