VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents

VAREX (VARied-schema EXtraction) is a benchmark for evaluating multimodal foundation models on structured data extraction from government forms. It comprises 1,777 documents with 1,771 unique schemas across three structural categories, each provided in four input modalities. Ground truth is deterministic — generated via a Reverse Annotation pipeline that programmatically fills PDF templates with synthetic values, validated through three-phase quality assurance achieving ~98.5% field-level accuracy.

Paper: arXiv:2603.15118

Evaluation code & scoring: github.com/udibarzi/varex-bench

Quick Start

from datasets import load_dataset
import json

ds = load_dataset("ibm-research/VAREX", split="benchmark")

doc = ds[0]
print(doc["doc_id"])        # e.g., "1044"
print(doc["split"])         # "Flat", "Nested", or "Table"

schema = json.loads(doc["schema"])
gt = json.loads(doc["ground_truth"])
image = doc["image"]        # PIL Image, 200 DPI
text = doc["text_layout"]   # Spatial text with layout

Columns

Column	Type	Description
doc_id	string	Unique document identifier
split	string	Structural category: Flat, Nested, or Table
image	Image	Document page rendered at 200 DPI (primary evaluation modality)
image_50dpi	Image	Document page rendered at 50 DPI (resolution robustness evaluation)
schema	string	JSON Schema defining the extraction target
ground_truth	string	JSON ground truth values
text_flow	string	Plain text in reading order
text_layout	string	Spatial text with whitespace-preserved layout

Input Modalities

Modality	Paper code	Column(s) to use
Plain Text	P	text_flow
Spatial Text	S	text_layout
Image	V	image (or image_50dpi for robustness)
Spatial Text + Image	S+V	text_layout + image

Document Splits

Split	Documents	Description
Flat	299	Simple key-value schemas, no nesting
Nested	1,146	Schemas with nested objects
Table	332	Schemas with arrays of objects

PDF Files

Original filled PDFs are available in the pdfs/ directory of this repository. Each filename corresponds to the doc_id column (e.g., doc_id "1044" → pdfs/1044.pdf). These allow researchers to apply their own text extraction or parsing pipelines.

Scoring

Evaluation code, scoring scripts, and field exclusion lists are maintained at: github.com/udibarzi/varex-bench

The benchmark uses Exact Match (EM) as the primary metric with order-invariant array matching via the Hungarian algorithm. 610 field-level exclusions are applied at scoring time for fields with known ground truth issues.

Citation

@inproceedings{varex2026,
  title   = {VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents},
  author  = {Barzelay, Udi and Azulai, Ophir and Shapira, Inbar and Friedman, Idan and Abo Dahood, Foad and Lee, Madison and Daniels, Abraham},
  year    = {2026}
}

License

Community Data License Agreement – Permissive, Version 2.0