NutriBench-Kitchen - Recipe Query (Hard): leaderboard
Recipe Query in NutriBench-Kitchen (Wei et al., 2026): questions on selecting or assessing a recipe under constraints from the observed inventory and nutritional knowledge, in the hard regime: different multi-turn questions without the frame hint. Score in percent (accuracy on multiple-choice questions, mean relative accuracy on numerical ones); higher is better.
Metric: Score (%; the recipe query task family, hard regime: different, multi-turn questions without the frame hint; NutriBench-Kitchen: 1,500 manually verified questions on 160 cooking videos (145 HD-EPIC segments and 15 self-recorded videos with measured ingredient quantities); accuracy on multiple-choice questions and mean relative accuracy on numerical ones; zero-shot with the same chain-of-thought prompt for every model). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 69.3 | #262 |
| 2 | Qwen 2.5 VL 7B Instruct | 60.4 | #1842 |
| 3 | GPT-4o | 50.5 | #741 |
| 4 | GLM-4.1V-9B (Thinking) | 33.7 | #1061 (GLM-4.1V-9B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=nutribench-kitchen-recipe-query-hard · How It Works · Data refreshed daily, snapshot 2026-11-12.