( fig.11 )
( fig.11 )
Andrew Luo, Jing Reyhan
June 9, 2026
Engineering
LongArray-Extract tests whether extraction systems can complete structured array outputs across hundreds or thousands of rows.
Jing Reyhan, Joseph Bajor, Cindy Hao
Engineering
RealDoc-Bench evaluates whether parsers preserve the structure agents need across real-world document workflows.
Joon Kim, Ameya Joshi, Cindy Hao, Jing Reyhan
Engineering
How Extend rebuilt its layout model for Parse 2.0, why layout detection drives parsing accuracy, and how stronger document structure improves deterministic pipelines, model routing, cost, and latency.
Jing Reyhan, Eli Badgio
Product
Today we're launching Parse 2.0, our SOTA layout-first document parsing API for agents, alongside RealDoc-Bench, an applied benchmark measuring parsing performance on the real-world documents agents actually encounter in production.
Jing Reyhan
Customers
How Flatiron Health replicated 6 months of in-house NGS extraction work in 2 weeks with Extend, scaling biomarker data across 5 million people with cancer.
Cindy Hao
Engineering
PoliTax Split evaluates document splitting on long public-sector tax packets with subtle document boundaries.
Joe Bajor, Cindy Hao
Customers
How Nuvocargo used Extend to hit 97-99% accuracy across document intake, classification, extraction, and shipment attribution, with near zero human involvement.
Cindy Hao
Customers
How Mercury uses Extend to power real-time document validation in onboarding, handling dozens of languages and formats with sub-7 second latency.
Cindy Hao
Engineering
Why we abandoned workflows for agents, how we learned to starve our context window, and why our optimization agent accidentally became a data quality tool.
Gus Eggert, Richard Li, Cindy Hao