Arrays
Array-heavy documents
Line items, repeated rows, and nested list extraction.
DocuPipe high setting30 documents
97.41%
Extendsame subset
91.98%
Document AI benchmark
Document AI rankings
This is Yardstick Lab's first live category. The current ranking covers a 50-document structured extraction run and includes DocuPipe product settings, Reducto, Extend, and direct model baselines.
Overall structured extraction score
Macro-average field accuracy from the current ranked dataset.
01
DocuPipe high settingbest overall score in the current run
97.56%
02
Direct model baseline Chighest model-only run
97.10%
03
DocuPipe standard settingstandard product setting for the same run
96.31%
04
Reducto Deep Extractagentic product setting
95.77%
05
Direct model baseline Amodel-only run without an extraction platform
94.64%
06
Direct model baseline Dadditional model-only run
93.68%
07
Reducto standardstandard product setting for the same run
92.70%
08
Extendscored product baseline
91.11%
09
Direct model baseline Bmodel-only run
90.53%
Direct model baselines are shown anonymized on this property; each is a frontier model called with no extraction platform. The per-model mapping is published in the underlying open benchmark.
50documents in the ranked dataset
6scored rows in the current leaderboard
3anonymous direct model baselines
1dated run behind this page
Slice rankings
Where the score separates
These slices come from the same current run and show evidence for specific buyer workflows.
Reconciliation
Reconciliation-heavy documents
Totals, consistency checks, and cross-field relationships.
DocuPipe high setting24 documents
96.89%
Extendsame subset
88.20%
Hard documents
Language stress cases
Right-to-left and CJK subsets from the current run.
DocuPipe high settingright-to-left subset
90.72%
Extendright-to-left subset
76.07%
DocuPipe high settingCJK subset
99.12%