DiningBench - Fat 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 fat; lower is better.

Metric: Mean absolute percentage error (%) of the estimated fat 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)26.53#123
2Gemini 3 Flash (Preview)27.09#156
3GPT-533.49#185
4O4 Mini37.8#372
5Gemma 3 12B (IT)41.12#1907
6Gemini 2.5 Pro42.19#280
7Qwen 3 VL 30B A3B Instruct43.37#903
8Qwen 3 VL 30B A3B (Thinking)43.66#727
9Qwen 3 VL 4B Instruct44.55#1348
10Gemini 2.5 Flash44.56#466
11Qwen 3 VL 8B Instruct48.02#1043
12GPT-4.149.29#506
13MiniCPM-V-4.549.69#1451
14Qwen 2.5 VL 32B Instruct50.05#1065
15Claude Sonnet 4.550.82#267

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=diningbench-fat-estimation · How It Works · Data refreshed daily, snapshot 2026-11-22.