DiningBench - Calorie 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; lower is better.
Metric: Mean absolute percentage error (%) of the estimated calories 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 19.99 | #123 |
| 2 | Gemini 3 Flash (Preview) | 20.57 | #156 |
| 3 | GPT-5 | 25.14 | #185 |
| 4 | O4 Mini | 27.68 | #372 |
| 5 | Gemma 3 12B (IT) | 27.95 | #1907 |
| 6 | Gemini 2.5 Pro | 28.61 | #280 |
| 7 | Gemini 2.5 Flash | 30.21 | #466 |
| 8 | Qwen 3 VL 8B Instruct | 31.06 | #1043 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 32.33 | #727 |
| 10 | Qwen 3 VL 30B A3B Instruct | 32.49 | #903 |
| 11 | Qwen 2.5 VL 32B Instruct | 32.66 | #1065 |
| 12 | MiniCPM-V-4.5 | 34.79 | #1451 |
| 13 | Claude Sonnet 4.5 | 35.01 | #267 |
| 14 | Qwen 2.5 VL 72B Instruct | 35.88 | #849 |
| 15 | GPT-4.1 | 36.88 | #506 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=diningbench-calorie-estimation · How It Works · Data refreshed daily, snapshot 2026-11-22.