Model releases, benchmarks, and research notes
How we build and evaluate the models behind Nanonets agents — from the Nanonets-OCR family to our work on complex instruction following.
A benchmark for whether AI can execute complex, multi-step enterprise processes correctly, from start to finish. We tested the whole field, and the best system gets only 38% of tasks fully right.
A deep dive into building fully automated document processing workflows — no human in the loop, no manual exceptions, no fallback queues.
The latest generation of our document OCR model — converting pages into structured, LLM-ready markdown with stronger layout, table, and equation understanding.
The second-generation Nanonets-OCR model, extending image-to-markdown extraction across more document types and languages.
Our open image-to-markdown OCR model — turning documents into structured text with tables, equations, and reading order preserved.
How Nanonets performs on Surge AI's ComplexConstraints benchmark for entangled instruction following — conditional, implicit, multistep, and negative constraints.
An open leaderboard tracking OCR model performance across document types, languages, and layout complexity — updated as new models are evaluated.
The industry benchmark for intelligent document processing — ranking models on real-world extraction tasks across invoices, forms, contracts, and more.
See it run on your process, with your documents.
Start free. No credit card. Or talk to our team about your workflow.