Databricks adds adaptive search model to speed agent retrieval

Databricks has enhanced its Adaptive Instructed-Retriever model to accelerate AI agent search tasks requiring multiple retrieval rounds, achieving response speeds twice as fast as comparable models like Claude Sonnet 5 and GPT-5.6 Luna. The model dynamically adjusts the number of search steps per query to balance speed and thoroughness, improving efficiency in complex information retrieval across large datasets.