DiningBench - Nutrition Estimation: leaderboard
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 February 2027. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 24.45 |
| 2 | Gemini 3 Flash (Preview) | 25.21 |
| 3 | GPT-5 | 32.17 |
| 4 | O4 Mini | 33.95 |
| 5 | Qwen 3 VL 30B A3B Instruct | 37.35 |
| 6 | Gemini 2.5 Pro | 38.21 |
| 7 | GPT-4.1 | 38.6 |
| 8 | Qwen 3 VL 8B Instruct | 39.24 |
| 9 | Qwen 2.5 VL 32B Instruct | 39.84 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 40.22 |
| 11 | Qwen 2.5 VL 72B Instruct | 40.56 |
| 12 | Gemini 2.5 Flash | 41.6 |
| 13 | GPT-4o | 42.43 |
| 14 | Qwen 3 VL 4B Instruct | 42.55 |
| 15 | Claude Sonnet 4.5 | 42.62 |
Interactive version: theaggregate.ai/benchmark?slug=diningbench-nutrition-estimation · How It Works · Data refreshed daily, snapshot 2026-10-07.