DocAttriBench - DocVQA: leaderboard

Metric: Overall answer correctness (%; grounded answer generation on DocVQA (467 test questions): an answer counts only when its text matches the reference and its evidence box is correct; zero-shot, with model-specific prompts and box parsers; element-level evidence boxes from the MAPPET attribution pipeline over Docling layout regions, a box counting when IoU >= 0.5). Source: arxiv.org. Saturation forecast: Around February 2028. 16 models tracked.

Top models

#ModelScore
1Qwen 3 VL 8B10.8
2Qwen 2.5 VL 32B Instruct7.8
3InternVL3-8B3.2
4InternVL3-38B3.2
5Qwen 2.5 VL 7B Instruct2.6
6InternVL2.5-2B0

Interactive version: theaggregate.ai/benchmark?slug=docattribench-docvqa · How It Works · Data refreshed daily, snapshot 2026-09-26.