TeachObs - Lesson Narrative Coverage: leaderboard
TeachObs lesson-level evaluation (Jeong et al., 2026): the model writes a narrative rating a whole classroom lesson in eight categories (design, delivery, learner response, materials, closure, strengths, areas for improvement, overall), and a GPT-5-mini judge checks which of the expert raters' atomic reference claims it covers. Coverage in percent, averaged over seven test lessons (1,030 claims); higher is better.
Metric: Reference coverage (%; share of the experts' atomic claims a lesson-level narrative covers, mean over seven test lessons, 1,030 claims, GPT-5-mini judge). Source: arxiv.org. Saturation forecast: Around 2029. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.3 (Low) | 33.1 | #332 (Grok 4.3) |
| 2 | Claude Opus 4.7 (Low) | 29.5 | #75 (Claude Opus 4.7) |
| 3 | Gemini 3.1 Pro (Preview) (Low) | 28.4 | #91 (Gemini 3.1 Pro (Preview)) |
| 4 | GPT-5.5 (Low) | 27.5 | #48 (GPT-5.5) |
| 5 | GLM-5V Turbo | 25.7 | #342 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=teachobs-lesson-narrative-coverage · How It Works · Data refreshed daily, snapshot 2026-11-12.