multimodal benchmark
DocVQA
A dataset for Visual Question Answering on document images containing 50,000 questions defined on 12,000+ document images. The benchmark tests AI's ability to understand document structure and content, requiring models to comprehend document layout and perform information retrieval to answer questions about document images.
Updated Aug 7, 2026
Published models26
Registry coverage26
MetricScore
EvidenceB
DocVQA leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
Score distribution
Top published rows on the benchmark's original scale.
DocVQA highlights
The top published results on this benchmark's own scale.
What is DocVQA?
Definition and scoring fields from the benchmark registry.
A dataset for Visual Question Answering on document images containing 50,000 questions defined on 12,000+ document images. The benchmark tests AI's ability to understand document structure and content, requiring models to comprehend document layout and perform information retrieval to answer questions about document images.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- DocVQA
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- docvqa|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about DocVQA.
Which model scores highest on DocVQA?
Qwen2.5 VL 72B Instruct is currently ranked first with 96.4%.
What does DocVQA measure?
A dataset for Visual Question Answering on document images containing 50,000 questions defined on 12,000+ document images. The benchmark tests AI's ability to understand document structure and content, requiring models to comprehend document layout and perform information retrieval to answer questions about document images.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
26 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.