Extend announced Operator-1, an autonomous document extraction agent that delivers state-of-the-art (SotA) accuracy on the most challenging extraction tasks. Operator-1 combines document processing tools with persistent memory in a dedicated workspace to extract data, validate its work, and reuse successful approaches. In evaluations, it set a new performance record across three document extraction benchmarks.

Complex documents challenge extraction systems that rely on chunking and a single model pass. Financial documents with long tables can contain skipped or duplicated rows, lengthy legal documents can scatter related information across thousands of pages, and dense diagrams can contain text that requires closer inspection. Operator-1 handles these tasks, so teams in critical industries can unlock data trapped in their documents and accelerate document processing with confidence.

Highest accuracy across three extraction benchmarks

Operator-1 sets a new standard for complex document extraction, achieving:

  • 99.3% on LongArray-Extract, an open-source extraction benchmark from Extend, which evaluates accurate and complete extraction of long arrays from real-world documents.

  • 99.88% on LongExtractionBench, a micro1 benchmark designed and commissioned by Reducto, which evaluates filling JSON schemas from long documents.

  • 96.57% on ExtractBench, a LlamaIndex benchmark, which evaluates structured extraction across business documents.

The results demonstrate saturation across three distinct extraction evaluations. Each benchmark uses a different dataset and evaluation methodology; two of the three benchmarks were published by third parties.

Key features

Operator-1 determines the steps each extraction task requires and works in a workspace designed to let models handle complex document extraction end-to-end:

  • Write and run code: Extract table rows, transform data, and validate calculations, such as comparing extracted line-item sums with a financial statement’s reported total.

  • Search across documents: Uses command-line and custom search tools to find and connect information across thousands of pages.

  • Crop and zoom: Inspect specific regions of document images, for instance looking at page breaks, and connect diagram details with surrounding text.

  • Validate and retry: Checks results against the task’s requirements and triggers another attempt when a check fails.

  • Learn from every run: Retain successful code and strategies in tenant-specific memory and runbooks for reuse on subsequent documents of the same type, reducing repeated work and potentially lowering runtime costs.

Pricing and availability

Operator-1 is available today in the Extend dashboard and API. It’s included on all Extend plans, including PAYG, Scale, and Enterprise. Operator-1 charges a fixed 5 credits per page.

To get started testing Operator-1 with 10,000 free credits, visit dashboard.extend.ai.

About Extend

Extend is a Document AI platform that provides models, APIs, and infrastructure to enable developers and agents to make every document process programmatic. Its end-to-end platform delivers API primitives and tooling to parse, extract, split, edit, run evals, and orchestrate workflows to turn unstructured documents into structured data that agents can reason over and act on. Companies including Zillow, Mercury, Opendoor, Brex, Amgen, First American Financial, C.H. Robinson, Flatiron Health, and more use Extend to power document ingestion in production.

Media gallery

About The Author
Business Wire