DiningBench - Fine-Grained Classification: 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 accuracy in percent at picking the dish among eight candidates from the same menu; higher is better.

Metric: Accuracy (%) choosing the dish among eight same-menu candidates (seven hard distractors) from multi-view user photos, over the 2,884 DiningBench classification samples; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 29 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash (Preview)81.83#156
2Gemini 3 Pro (Preview)81.55#123
3Gemini 2.5 Pro73.51#280
4Gemini 2.5 Flash71.01#466
5GPT-570.18#185
6GPT-4.168.59#506
7Qwen 3 VL 30B A3B Instruct65.43#903
8Qwen 2.5 VL 72B Instruct65.29#849
9GPT-4o65.26#754
10O4 Mini64.81#372
11Qwen 3 VL 8B Instruct64.15#1043
12Qwen 3 VL 30B A3B (Thinking)61.34#727
13Qwen 2.5 VL 32B Instruct61.17#1065
14Qwen 2.5 VL 7B Instruct60.85#1919
15Qwen 3 VL 4B Instruct60.06#1348

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

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