M3-VQA (Gold Section Evidence): leaderboard

M3-VQA (CASIA, 2026) is a knowledge-based visual question answering benchmark of multimodal, multi-entity, multi-hop questions that require recognizing several entities and chaining evidence from Wikipedia, scored by the overlap between predicted and gold answer sets. This board is IoU accuracy in percent with the Wikipedia sections holding the gold evidence of every hop added to the input; higher is better.

Metric: IoU accuracy (%) between the predicted and gold answer sets (exact match against Wikidata answer aliases, one-year margin for dates), averaged over the M3-VQA multi-entity, multi-hop knowledge-based visual questions, with the Wikipedia sections containing the gold evidence of every hop added to the input; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL2.5-78B55.16#1031
2InternVL2.5-38B53.95#1373
3GPT-4o53.07#751
4Qwen 2.5 VL 72B Instruct52.35#841
5Qwen 2.5 VL 32B Instruct50.02#1062
6InternVL2.5-26B45.5#2156
7LLaVA-OneVision-7B43.62#2981
8Qwen 2.5 VL 7B Instruct42.66#1902
9InternVL2.5-8B42.12#2370
10Qwen 2.5 VL 3B Instruct39.69#2713
11InternVL2.5-4B39.34#2549
12MiniCPM-V-2.639.11#3267
13Qwen 2 VL 7B Instruct38.92#2573
14DeepSeek-VL238.81#2927
15InternVL2.5-2B31.94#6468

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

Interactive version: theaggregate.ai/benchmark?slug=m3-vqa-gold-section-evidence · How It Works · Data refreshed daily, snapshot 2026-11-21.