DiningBench - Nutrition Estimation: leaderboard

DiningBench (2026) evaluates vision-language models on multi-view user photos of restaurant dishes, testing dish classification against same-menu distractors, nutrition estimation and food visual question answering. This board is the mean absolute percentage error of the estimated calories, protein, carbohydrates and fat, averaged over the four nutrients; lower is better.

Metric: Mean absolute percentage error (%) of the estimated calories, protein, carbohydrates and fat (mean of the four MAPEs) over the 1,650 DiningBench nutrition-estimation dishes (multi-view restaurant photos; ground truth from merchant labels or ingredient-based estimates cross-checked against USDA FoodData Central); lower is better. Source: arxiv.org. Saturation forecast: Around March 2027. 29 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview)24.45#119
2Gemini 3 Flash (Preview)25.21#152
3GPT-532.17#178
4O4 Mini33.95#368
5Qwen 3 VL 30B A3B Instruct37.35#895
6Gemini 2.5 Pro38.21#276
7GPT-4.138.6#505
8Qwen 3 VL 8B Instruct39.24#1029
9Qwen 2.5 VL 32B Instruct39.84#1062
10Qwen 3 VL 30B A3B (Thinking)40.22#719
11Qwen 2.5 VL 72B Instruct40.56#841
12Gemini 2.5 Flash41.6#463
13GPT-4o42.43#751
14Qwen 3 VL 4B Instruct42.55#1332
15Claude Sonnet 4.542.62#261

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=diningbench-nutrition-estimation · How It Works · Data refreshed daily, snapshot 2026-11-21.