Introducing Batch Cortex Search: Hybrid Search Engine for High-Throughput Workloads

Snowflake introduces Batch Cortex Search, a function that enables high-throughput, offline hybrid search (vector + keyword) by submitting batches of queries to an existing Cortex Search Service. While standard Cortex Search is optimized for low-latency interactive queries, Batch Cortex Search is designed for workloads requiring thousands of simultaneous searches — such as entity resolution, deduplication, catalog mapping, audience matching, content tagging, and similar item discovery. At 10,000 queries, Batch Cortex Search is 47.5x faster than the standard service due to parallelism. Setup requires only a few lines of SQL or Python, with no need to manage embeddings or vector databases.