VLR-CVC/DocVQA-2026

DocVQA2026 | ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains

Building upon previous DocVQA benchmarks, this evaluation dataset introduces challenging reasoning questions over a diverse collection of documents spanning eight domains, including business reports, scientific papers, slides, posters, maps, comics, infographics, and engineering drawings.

By expanding coverage to new document domains and introducing richer question types, this benchmark seeks to push the boundaries of multimodal reasoning and promote the development of more general, robust document understanding models.

Results

Category	Gemini 3 Pro Preview	GPT-5.2	Gemini 3 Flash Preview	GPT-5 Mini
Accuracy	0.375	0.350	0.3375	0.225
Business Report	0.4	0.6	0.2	0.3
Comics	0.3	0.2	0.4	0.1
Engineering Drawing	0.3	0.3	0.5	0.2
Infographics	0.7	0.6	0.5	0.5
Maps	0.0	0.2	0.0	0.1
Science Paper	0.3	0.4	0.5	0.1
Science Poster	0.3	0.0	0.2	0.0
Slide	0.7	0.5	0.4	0.5

• GPT Models: "High thinking" and temperature of 1
• Gemini Models: "High thinking" and temperature of 0

Both models where evaluated via their respective APIs. If one sample fails because the input files are too large, the result counts of a failure (e.g. the limit for file inputs for OpenAI models is 50MB, we have some comics that surpass that). Planning on realising new eval results in the coming days.

Dataset Structure

The dataset consists of:

1. Images: High-resolution PNG renders of document pages located in the images/ directory.
2. Annotations: A Parquet file (val.parquet) containing the questions, answers, and references to the image paths.

Schema

Field	Type	Description
id	string	Unique identifier for the question (e.g., slide_1_q1).
doc_id	string	Identifier for the document context.
image_paths	list[string]	A list of relative paths to all pages associated with the document.
question	string	The visual question.
answer	string	The ground truth answer.
doc_category	string	The type of document (e.g., science_paper, financial_report).