AI Search High QPS
GALets you set a target queries per second on a standard AI Search endpoint, and Databricks provisions extra capacity to reach it on a best-effort basis; without it a standard endpoint serves roughly 20 to 200 QPS depending on index size. You set the target under Advanced Settings in the endpoint UI, or with target_qps in the Python SDK or REST API. The extra capacity is billed whether or not traffic uses it, it does not autoscale, storage-optimized endpoints are not supported, and only queries sent to the index URL with service principal OAuth get the higher throughput. It was documented as a Beta and then a Public Preview under the Vector Search name before it became generally available on July 15, 2026.
Status history
- GA
Sources (2)
- 2026-10-27Databricks docs GA
- 2026-07-15Release notes GA
Get Tuesday's version of this
What shipped in Databricks and why it matters, once a week. Free, one-click unsubscribe.
Statuses and dates come from public Databricks sources only. We do not see the private roadmap Databricks shares with customers. If something looks wrong, report a correction.