Zilliz, the company behind the open-source vector database Milvus, is debuting the latest iteration of Milvus, which unites hybrid semantic and keyword search into a single, high-performance platform. This centralization dramatically boosts query processing speed—30x faster than traditional solutions—driving accessibility of AI-powered, advanced search to a variety of enterprises, according to Zilliz.
At its core, Milvus 2.5 forwards Zilliz’s mission of wanting “to help developers,” according to Charles Xie, founder and CEO of Zilliz. “We want to help every single company on this planet to process unstructured data at a larger scale, in a fast fashion, and process unstructured data more efficiently.”
In this release, developer efficiency is improved by merging semantic and keyword search into a single system, rivalling traditional methods where hybrid search is forced into two separate systems. When semantic and keyword search are maintained in two different locations, a litany of operational inefficiency occurs due to duplicate infrastructures that force complex integrations, wasted resources, and increased costs, according to Zilliz.
"We're witnessing a fundamental shift in how organizations approach search," said Xie. "With Milvus 2.5, we're not just combining two search approaches—we're revolutionizing enterprise search with a solution that's 30 times faster while dramatically simplifying infrastructure. This is a game-changer for organizations building AI-powered applications."
Milvus 2.5’s intelligent query processing enables one request to execute both semantic and full-text search, reducing the API calls needed to separate systems. Paired with Sparse-BM25 technology—a sparse vector implementation of the BM25 algorithm—Milvus 2.5 can return search results in just 6 milliseconds with 1 million vectors on fully managed Zilliz Cloud.
Compared to Elasticsearch—which takes 200 milliseconds when tested on fully managed Elastic Cloud with the same parameters—Milvus 2.5 delivers ultra-fast search underscored by eased maintenance and consistent accuracy as document collections grow and evolve, according to Zilliz.
Because Milvus 2.5 consolidates hybrid search into a single system, data management is holistically streamlined, reducing inefficiencies. A unified table structure stores dense (vector-based) and sparse (keyword-based) data alongside shared metadata labels, preventing metadata deduplication. Atop of this efficiency, Milvus 2.5’s centralized approach extends to security and access control, where, with only one system to maintain, access controls, compliance, and consistency are fortified.
To learn more about Milvus 2.5, please visit https://zilliz.com/.