At Datalab, we train models that are fast and state of the art.  Our main model, Surya, is only 800M parameters, but tops several text extraction benchmarks.  We use this model, along with a lot of logic, to extract and format text.  In most cases, it works extremely well.

But documents always have edge cases.  Tables with scribbles on them, handwritten redlines on legal documents, or visually dense forms.  Sometimes you don't want the exact text that's on the page - you want a cleaned up representation that's easy to understand.  This is why we're introducing High AccuracyMode, which blends our own models, including larger models that we've trained, with frontier LLMs to format content optimally.

High Accuracy mode means that you don't have to try to figure out what the document is saying - it's automatically inferred and simplified for you.  Check out some of the examples below - we think you'll be excited about the results.

High Accuracy Mode is Now Live!

High Accuracy mode is live and available in Forge and in our public Playground (limit to 1 page).

Pages processed using High Accuracy mode are billed at $6/1000 pages.