Founder Brian Raymond says the rebuilt parsing pipelines beat three frontier models on tables in a 224-page company test, though the methodology remains vendor-reported.
By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)
· Published
Primary source: [Aligned News - AI Intelligence](https://x.com/_Brian_Raymond/status/2112993839782973685)
Why it matters #
Unstructured is selling a specialized parsing step for agent workflows where a misplaced table value can change a downstream decision. Its benchmark favors Transform on tables, but the sample and scoring rules are company-reported, so customers still need to test it on their own documents.
Unstructured has upgraded Transform, its document-parsing product, with rebuilt pipelines and models aimed at keeping tables, forms and scans usable in AI workflows. Founder and CEO Brian S. Raymond (@_Brian_Raymond) framed the October 21st release around a practical problem: a model can read the words on a page and still attach a number to the wrong row.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.