Running Open-Jev in SQL on Databricks
Databricks, Thursday, September 24th, 2026
Databricks showed how to run open-weight 'decision model' Jev variants like SemIf-OpenJev directly from SQL using serverless GPUs.
Databricks describes running open-weight 'System One' decision models, a class of foundation models such as Jev that produce fast, cheap, well-calibrated decisions from a discrete set of options, directly on governed Databricks data.
Using serverless GPUs and managed Model Serving, teams can query custom model endpoints from SQL or Lakeflow jobs with the ai_query function, returning structured decisions and probability scores at scale.
The post walks through an importable notebook that deploys an open-source model called SemIf-OpenJev to classify hotel reviews as good or bad, using Databricks AI Runtime to avoid managing GPU infrastructure directly.