DiningBench - Carbohydrate Estimation: leaderboard
Metric: Mean absolute percentage error (%) of the estimated carbohydrates 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) | 27.4 |
| 2 | Gemini 3 Flash (Preview) | 27.94 |
| 3 | GPT-4.1 | 35.81 |
| 4 | GPT-5 | 36.63 |
| 5 | O4 Mini | 37.38 |
| 6 | GPT-4o | 40.19 |
| 7 | Qwen 3 VL 30B A3B Instruct | 41.53 |
| 8 | Qwen 2.5 VL 72B Instruct | 41.7 |
| 9 | Gemini 2.5 Flash | 42.62 |
| 10 | Gemini 2.5 Pro | 43.01 |
| 11 | Qwen 2.5 VL 32B Instruct | 43.92 |
| 12 | Qwen 3 VL 8B Instruct | 43.99 |
| 13 | GPT-4o Mini | 45.03 |
| 14 | Qwen 3 VL 30B A3B (Thinking) | 48.12 |
| 15 | Qwen 3 VL 4B Instruct | 49.28 |
Interactive version: theaggregate.ai/benchmark?slug=diningbench-carbohydrate-estimation · How It Works · Data refreshed daily, snapshot 2026-10-07.