SIT-Eval - Specialized Tutors - Overall: leaderboard
SIT-Eval (2026), simulated adaptive tutoring over the 100 GSM8K problems specified for the scored benchmark in section 4.3 (section 4.1 instead says 170, an unresolved source inconsistency), twelve rules aggregated into seven categories; LLM-judged rule adherence, temperature 0.8 and 1024 output tokens. Specialized Tutors protocol. Specialized-table Gemini row is not admitted until the source resolves its conflict with the prose LearnLM identity. Rule Adherence (0–1); higher is better.
Metric: Rule Adherence (0–1). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen-2-5-7B-SocraticLM | 0.64 | |
| 2 | Llemma-7B-Science-Tutor | 0.29 | |
| 3 | Qwen-2-5-Math-7B-Instruct | 0.17 |
Interactive version: theaggregate.ai/benchmark?slug=sit-eval-specialized-tutors-overall · How It Works · Data refreshed daily, snapshot 2026-10-20.