Introducing balanced extraction mode

A new mode that trades a little speed for higher recall on hard fields.

What's new

This update centers on Docen Extract. The short version: cleaner output on the documents that used to give it trouble, and fewer surprises on everything else.

The improvements come from better training data, a sharper decoding pass, and evaluation that caught regressions we would otherwise have shipped.

• Improved handling of recall.
• Improved handling of accuracy modes.
• Improved handling of hard fields.

By the numbers

Getting started

The update is live on the managed API and available for self-hosted deployments. Existing integrations pick it up with no code changes — the output format is unchanged.

Want to see it on your own documents? Open the playground or reach out — we're happy to run a sample with you.