M3-VQA (Gold Sentence 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 gold evidence sentence of every reasoning 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 gold evidence sentences of every reasoning hop from the linked Wikipedia pages 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
1InternVL2.5-78B58.71#1031
2GPT-4o58.63#751
3InternVL2.5-38B58.51#1373
4Qwen 2.5 VL 72B Instruct58.41#841
5Qwen 2.5 VL 32B Instruct53.48#1062
6Qwen 2.5 VL 7B Instruct48.97#1902
7InternVL2.5-26B48.5#2156
8LLaVA-OneVision-7B47.15#2981
9MiniCPM-V-2.646.92#3267
10InternVL2.5-8B46.82#2370
11DeepSeek-VL244.64#2927
12InternVL2.5-4B42.77#2549
13Qwen 2.5 VL 3B Instruct42.6#2713
14Qwen 2 VL 7B Instruct41.21#2573
15InternVL2.5-2B34.47#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-sentence-evidence · How It Works · Data refreshed daily, snapshot 2026-11-21.