Snowflake raises AI_EXTRACT page limit, with cross-region inference required
Snowflake’s November 2 release notes describe a Public Preview update to AI_EXTRACT, its document-extraction function. The update increases the document page limit and changes how extraction handles complex layouts. It is not a general-availability announcement.
The maximum document size rises from 125 pages to 1,000 pages. Snowflake says the update can extract fields, lists and tables from documents such as forms, handwritten content, dense tables and charts.
Snowflake reports a 12.8% relative improvement in average extraction accuracy compared with the previous version in its own evaluations. That is a vendor evaluation result, not an independent benchmark or a guarantee for every document.
The update also aims to handle requests with many fields more reliably and return one combined result. Snowflake says users no longer need to adjust scale_factor manually: existing calls still work, but any value supplied for that parameter is ignored.
Cross-region inference must be enabled to receive the update. Snowflake says the changes then apply automatically, without SQL changes or a price change. Accounts without cross-region inference continue to use the previous version.
Teams should check their cross-region inference setting before expecting the larger page limit or new extraction behavior. The stated lack of a price change concerns this update; it does not mean AI_EXTRACT is free or that every workload will have the same total cost.