ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains

Artemis Llabrés; Marc Serra Ortega; Tomàs Ockier; Samuel Ortega Cuadra; Amritpal Singh; Christos Georgakilas; Andrey Barsky; Ernest Valveny

cs.CLSep 8, 2026

ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains

Authors: Artemis Llabrés, Marc Serra Ortega, Tomàs Ockier, Samuel Ortega Cuadra, Amritpal Singh, Christos Georgakilas, Andrey Barsky, Ernest Valveny,

Organizations: Computer Vision Center and Universitat Autònoma de Barcelona

Abstract

In this report we present results of the ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple Domains. This competition aimed to advance research in document understanding through the task of Visual Question Answering (VQA). Building upon previous DocVQA benchmarks, this competition 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. The competition concluded with 20 valid submissions from 8 teams spanning zero-shot VLMs, OCR and parser-augmented pipelines, agentic retrieval systems, multi-agent ensembles, and fine-tuned multimodal models. The results show that the strongest systems move beyond single-pass prompting and instead rely on structured evidence extraction, retrieval, verification, and orchestration across multiple components.

Figures & tables

Figure 1: Evolution of submitted methods to the DocVQA competitions.

Figure 2: Overview of the eight document domains of the competition: business reports, scientific publications, slides, posters, maps, comics, infographics, and engineering drawings.

Figure 3: Sample question from the maps domain: to answer it, the model must locate stand 30, determine the map’s north orientation, follow the main avenue northward to its endpoint, and then use the legend to translate the final stand number into its corresponding name.

Figure 4: Sample question from the comics domain: to answer it, the model must identify the page where the murder occurs, track the sequence of locations mentioned in the narrative panels leading up to that event, and determine the last state referenced before the state in which the crime is committed.

Figure 5: Questions per Document Distribution (Validation+Test)

Panel of Figure 6 — Figure 6: Prompt used for all baseline models.

Cat.	Method	Type	Accuracy
$>$ 35B	Luna + IDA + Multi-Agent Ensemble + with FT	Multi-agent ensemble	60.00
$>$ 35B	Luna + IDA + Multi-Agent Ensemble	Multi-agent ensemble	55.63
$>$ 35B	Uni-Parser Tools + Gemini-3.1-Pro	Parser/OCR zero-shot	46.25
$>$ 35B	ARGUS_Gemini_3.1_Pro	Agentic RAG	43.75
$>$ 35B	Gemini-3.1-Pro-preview	Single-model zero-shot	37.50
$>$ 35B	Gemini-3-Flash-preview	Single-model zero-shot	35.63

Table 1: Overall DocVQA competition results across all submitted methods, grouped in a single table spanning the three parameter bands (

\leq

8B, 8–35B, and

>

35B). Accuracy is shown on a 0–100 percentage scale. Boldface indicates the winner in each parameter band, with a tie in the 8–35B band. Light-gray rows correspond to official baselines.

Figure 7: Domain-wise accuracy for selected representative methods. The figure highlights differences in difficulty across domains and shows that the strongest systems are more balanced than the zero-shot baselines across categories.

Figure 8: Domain-wise comparison of the two tied methods in the 8–35B category. Although ARGUS_Qwen3.5_27B and Perceive-Reason-Code achieve the same overall accuracy, they do so through different domain strengths.

Document Understanding

Document understanding enables machines to interpret text, layout, tables, and images to extract information and answer questions about documents.

See all 93 papers

Doc-V*:Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA
Apr 15, 2026Yuanlei Zheng, Pei Fu, Hang Li +9Document QA Long-Document VQA
School of Software Engineering, Huazhong University of Science and Technology · MiLM Plus, Xiaomi Inc. · School of Computer Science and Technology, Huazhong University of Science and Technology +1
Comparative Study of Domain-adapted VLMs for General Document Visual Question Answering
Jul 8, 2026Miguel Lopez-Duran, Elena Marrero, Julian Fierrez +8VLM Evaluation Document Understanding
BiometricsAI, Universidad Autónoma de Madrid (UAM), Spain
AutothinkRAG: Complexity-Aware Control of Retrieval-Augmented Reasoning for Image-Text Interaction
Mar 5, 2026Jiashu Yang, Chi Zhang, Abudukelimu Wuerkaixi +5Multimodal RAG Multimodal QA
Dalian University of Technology · Meituan LongCat Interaction Team · Tsinghua University

Apr 15, 2026J/K move · Enter open · S save

Doc-V*:Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA

Yuanlei Zheng, Pei Fu, Hang Li, Ziyang Wang+8

School of Software Engineering, Huazhong University of Science and Technology · MiLM Plus, Xiaomi Inc. · School of Computer Science and Technology, Huazhong University of Science and Technology+1

Document QA Long-Document VQA VLM Reasoning

Open paper AskPDF arXiv

Multi-page Document Visual Question Answering requires reasoning over semantics, layouts, and visual elements in long, visually dense documents. Existing OCR-free methods face a trade-off between capacity and precision: end-to-end models scale poorly with document length, while visual retrieval-based pipelines are brittle and passive. We propose Doc-

V^*

, an OCR-free agentic framework that casts multi-page DocVQA as sequential evidence aggregation. Doc-

V^*

begins with a thumbnail overview, then actively navigates via semantic retrieval and targeted page fetching, and aggregates evidence in a structured working memory for grounded reasoning. Trained by imitation learning from expert trajectories and further optimized with Group Relative Policy Optimization, Doc-

V^*

balances answer accuracy with evidence-seeking efficiency. Across five benchmarks, Doc-

V^*

outperforms open-source baselines and approaches proprietary models, improving out-of-domain performance by up to 47.9% over RAG baseline. Other results reveal effective evidence aggregation with selective attention, not increased input pages.

Abstract

Document Understanding

Explore similar work

Doc-V*:Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA

Comparative Study of Domain-adapted VLMs for General Document Visual Question Answering

AutothinkRAG: Complexity-Aware Control of Retrieval-Augmented Reasoning for Image-Text Interaction