Who leads the FUNSD benchmark?

LayoutLMv3-large currently leads FUNSD with a score of 92.08 on f1.

What is the state-of-the-art score on FUNSD?

The state-of-the-art result on FUNSD is 92.08 (f1), achieved by LayoutLMv3-large as of 2026.

How many models are tracked on FUNSD?

Codesota tracks 13 models on FUNSD.

When was the FUNSD leaderboard last updated?

The FUNSD leaderboard on Codesota includes results through 2026.

Codesota · Benchmark · FUNSDHome/Leaderboards/FUNSD

Unknown

FUNSD.

Name: FUNSD Benchmark Results
Creator: Unknown
Published: 2026-01-01
License: https://creativecommons.org/licenses/by/4.0/

199 fully annotated forms. Tests semantic entity labeling and linking.

Paper ↗Leaderboard ↓Lineage

§ 01 · Leaderboard

Results by metric.

Found a wrong score or missing run?

Use row edits to send a sourced correction into moderation.

Add / edit result ↗Report issue ↗

f1

F1 is the reported evaluation metric for FUNSD. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for f1verifiedpapervendorcommunityunverified

Muted rows were not state of the art when published — an earlier or same-year result already scored better.

Rank	Model	Trust	Score	Year	Links	Fix
01	LayoutLMv3-large Fetched from CodeSOTA API on 2026-04-20	verified	92.08	2026	Source ↗	Looks wrong?
02	UDOP Fetched from CodeSOTA API on 2026-04-20	verified	91.62	2026	Source ↗	Looks wrong?
03	LayoutLMv3-base Fetched from CodeSOTA API on 2026-04-20	verified	90.29	2026	Source ↗	Looks wrong?
04	DocFormerv2-large Fetched from CodeSOTA API on 2026-04-20	verified	88.89	2026	Source ↗	Looks wrong?
05	LiLT[EN-R2]-base Fetched from CodeSOTA API on 2026-04-20	verified	88.41	2026	Source ↗	Looks wrong?
06	DocFormerv2-base Fetched from CodeSOTA API on 2026-04-20	verified	88.37	2026	Source ↗	Looks wrong?
07	StructuralLM Fetched from CodeSOTA API on 2026-04-20	verified	85.14	2026	Source ↗	Looks wrong?
08	FormNet Fetched from CodeSOTA API on 2026-04-20	verified	84.69	2026	Source ↗	Looks wrong?
09	BROS-large Fetched from CodeSOTA API on 2026-04-20	verified	84.52	2026	Source ↗	Looks wrong?
10	LayoutLMv2-large Fetched from CodeSOTA API on 2026-04-20	verified	84.2	2026	Source ↗	Looks wrong?
11	LayoutLMv2-base Fetched from CodeSOTA API on 2026-04-20	verified	82.76	2026	Source ↗	Looks wrong?
12	LayoutLMv1-base Fetched from CodeSOTA API on 2026-04-20	verified	79.27	2026	Source ↗	Looks wrong?
13	LayoutLMv1-large Fetched from CodeSOTA API on 2026-04-20	verified	77.89	2026	Source ↗	Looks wrong?

Lineage

FUNSD in context.

See full ocr benchmarks lineage →

Predecessors (1)

active2002-09

IAM

From transcription to structure: FUNSD reframed OCR as 'find the question, link to its answer' rather than 'recognise every character'. The shift that produced LayoutLM and the entire form-understanding line.

This benchmark (1)

saturated2019-05

FUNSD

Successors (1)

superseded2023-05

OCRBench

Once VLMs could read at all, evaluation needed to span more than forms. OCRBench bundled scene text, document VQA, KIE and handwritten math into one composite — the first VLM-era OCR benchmark.

§ 04 · Submit a result

Add to the leaderboard.

Submit a Result

← Back to Leaderboards