_
Measured on long, real-world documents. We publish results, methodology, and source data so you can verify them yourself.
_
Public benchmarks for document parsing, extraction, and splitting, with published methodology and source data.
Parsing
Measures whether parsers deliver accurate layouts, preserve reading order, and enable agents to correctly answer objective questions against real-world documents.
Q&A accuracy
Extend Parse 2.0
95.7%
Layout F1
Adjusted F1
0.847
Q&A set
Prompts across 581 docs
1,359
Extracting
Tests whether extraction systems preserve cardinality and return complete, schema-faithful arrays when the output grows from a dozen of rows to thousands.
Mean accuracy
Extend across 45 PDFs
99.2%
Run completion
45 of 45 PDFs completed
100%
Largest array
Rows in one PDF
2.2k
Splitting
Evaluates document splitting on long, compound tax filings where frontier models miss subtle boundaries across hundreds of pages.
Best harness F1
Claude Opus 4.6
72.48%
Recall lift
Points across models
17-44
Evaluation set
Largest PoliTax docs
30
( fig.11 )