DiningBench - VQA: 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 on food visual questions about cooking technique, dietary advice and counterfactual or multi-image reasoning, judged against references; higher is better.

Metric: Accuracy (%) on the 804 DiningBench visual questions (cuisine technique, dietary advice, counterfactual and multi-image reasoning), answers judged by an LLM against the reference; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 29 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview)90.42#119
2Gemini 2.5 Pro89.93#276
3Gemini 3 Flash (Preview)88.56#152
4GPT-586.94#178
5GPT-4.184.33#505
6Claude Sonnet 4.583.58#261
7GPT-4o80.6#751
8Qwen 3 VL 30B A3B Instruct80.6#895
9O4 Mini80.35#368
10Gemini 2.5 Flash78.36#463
11Qwen 2.5 VL 72B Instruct76.62#841
12MiMo-VL-7B-RL75#1155
13Qwen 3 VL 30B A3B (Thinking)74.13#719
14Qwen 3 VL 8B Instruct72.76#1029
15InternVL3.5-38B72.51#1021

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=diningbench-vqa · How It Works · Data refreshed daily, snapshot 2026-11-21.