DiningBench - Protein Estimation: leaderboard
Metric: Mean absolute percentage error (%) of the estimated protein 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 May 2027. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 23.88 |
| 2 | Gemini 3 Flash (Preview) | 25.23 |
| 3 | Qwen 3 VL 30B A3B Instruct | 32.01 |
| 4 | GPT-4.1 | 32.42 |
| 5 | Qwen 2.5 VL 32B Instruct | 32.73 |
| 6 | O4 Mini | 32.96 |
| 7 | Claude Sonnet 4.5 | 33.15 |
| 8 | Qwen 2.5 VL 72B Instruct | 33.38 |
| 9 | GPT-5 | 33.4 |
| 10 | Qwen 3 VL 8B Instruct | 33.87 |
| 11 | Qwen 2.5 VL 7B Instruct | 34.67 |
| 12 | GPT-4o Mini | 35.08 |
| 13 | Qwen 3 VL 4B Instruct | 36 |
| 14 | GPT-4o | 36.03 |
| 15 | Qwen 3 VL 30B A3B (Thinking) | 36.77 |
Interactive version: theaggregate.ai/benchmark?slug=diningbench-protein-estimation · How It Works · Data refreshed daily, snapshot 2026-10-07.