TeachObs - Scene Coding (Transcript + Frame): leaderboard
The TeachObs scene-coding task (Jeong et al., 2026) with the scene's mid-frame image attached to its transcript: the model marks which of 39 teaching-observation codes apply to each 15-second classroom scene. Macro F1 in percent over the codes against trained coders' gold labels, 1,099 scenes of six test lessons; higher is better.
Metric: Macro F1 (%; 39 observation codes per 15-second classroom scene from its transcript and mid-frame image, 1,099 scenes of six test lessons). Source: arxiv.org. Saturation forecast: Around June 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (Low) | 51 | #91 (Gemini 3.1 Pro (Preview)) |
| 2 | Claude Opus 4.7 (Low) | 43.8 | #75 (Claude Opus 4.7) |
| 3 | GPT-5.5 (Low) | 43.5 | #48 (GPT-5.5) |
| 4 | Grok 4.3 (Low) | 41.3 | #332 (Grok 4.3) |
| 5 | GLM-5V Turbo | 37.8 | #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-scene-coding-transcript-plus-frame · How It Works · Data refreshed daily, snapshot 2026-11-12.