StudentBench - Practice Design: leaderboard
Expert human tutors compare pairs of GRE practice problems written by AI tutors on three criteria; preferences are fitted to a Bradley-Terry score centred on zero. Each row is one model at one reasoning setting. Higher is better. Source: studentbench.org.
Metric: Expert tutor pairwise preference, Bradley-Terry score (practice problems). Source: studentbench.org. Saturation forecast: Around November 2027. 12 models tracked. Updated 4 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Opus 5 | High | 1.46 | #26 |
| 2 | Claude Opus 4.8 | xHigh | 1.14 | #64 |
| 3 | GPT-5.5 Pro | Medium | 0.93 | #40 |
| 4 | GPT-5.5 | High | 0.87 | #54 |
| 5 | Kimi K2.6 | Medium | 0.11 | #192 |
| 6 | Claude Sonnet 4.6 | Low | -0.04 | #175 |
| 7 | Claude Sonnet 5 | Low | -0.26 | #118 |
| 8 | Gemini 3.1 Pro (Preview) | High | -0.55 | #101 |
| 9 | Gemini 3.7 Flash | Medium | -0.62 | #42 |
| 10 | Gemini 3.5 Flash | Low | -1.3 | #108 |
| 11 | GPT-5.4 Mini | Non-reasoning | -1.31 | #405 |
| 12 | Gemma 4 31B | High | -1.4 | #389 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=studentbench-practice-design · How It Works · Data refreshed daily, snapshot 2026-11-20.