SpecialEduBench - Attitude: leaderboard
The attitude axis of SpecialEduBench (Na et al., 2026): 68 items built on recorded intervention test honesty and professional dispositions, such as reporting the progress of a child truthfully when a caregiver asks otherwise, crossed with pressure and monitoring conditions. An LLM judge rates each response on a 1 to 10 rubric, and the board is the mean over all responses; higher is better.
Metric: Mean rubric score on a 1 to 10 scale over all attitude responses (68 items crossed with pressure and monitoring conditions into 192 cells, ten draws per cell on the two honesty families; judged at temperature 0 against the rubric by an LLM judge aligned to expert reference scores (the paper names gemma-4-31b-it and qwen3.6-27b as its two judges and reports the second only in the text)). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Low) | 9.32 | #20 (Claude Opus 5) |
| 2 | Grok 4.5 | 8.36 | #78 |
| 3 | Gemini 3.1 Pro (Preview) | 7.36 | #91 |
| 4 | Mistral Large 3 | 7.28 | #819 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 7.11 | #451 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=specialedubench-attitude · How It Works · Data refreshed daily, snapshot 2026-11-12.