We’re excited to introduce new capabilities and performance improvements for vector and full-text search in Azure Cosmos DB. These updates make it easier for developers and architects to build semantic search experiences and generative AI applications on the world’s most scalable NoSQL database.

Note: These new capabilities are actively rolling out. As a result, they may not yet be available in all Azure regions.

Float16 Vector Support

Save up to 50% on storage while maintaining high recall. You can now store and index vectors in Float16 format, reducing storage costs and memory usage while keeping recall accuracy high. This is useful for large datasets or multi-tenant systems with a significant number of embeddings and optimizing storage costs is important. Learn more.

Example: A vector embedding policy now supports float16 as an option in the dataType parameter:

{
    "vectorEmbeddings": [
        {
            "path":"/vector1",
            "dataType":"float16",
            "distanceFunction":"cosine",
            "dimensions":1536
        }
    ]
}

Faster vector inserts, indexing, and filtered search.

The latest Azure Cosmos DB engine optimizations deliver higher throughput and lower latency for vector insert and update operations. Developers running large ingestion pipelines or applying real-time filters across partitions will see measurable gains in both speed and consistency. Up to a 30% improvement in vector index ingestion throughput and ~50% reduction in P50 latency compared to our previous performance. Vector search performance has also improved, with 300% lower P99 latency on vector searches and 32% higher recall for filtered searches using Sharded DiskANN.

These gains help developers handle larger datasets and higher query volumes with lower RU consumption and more predictable performance. Over the next few months, additional optimizations for indexing speed and parallel build performance will roll out globally, further reducing time-to-ready for large collections and accelerating workloads that rely on frequent vector updates.

Full-text search: Fuzzy Search (GA)

Fuzzy search is now generally available, allowing your applications to handle misspellings, plurals, and near matches without custom logic. This makes text retrieval more resilient and user-friendly for agents, search experiences, and other workloads where finding relevant text is critical. Learn more.

Example

The following query returns documents containing words close to “microsoft”. So, documents with the properly spelled “microsoft” would appear, as well words that have a distance of 2 modifications, such as “microosoft” and “microosofft”

SELECT TOP 10 *
FROM c
WHERE FullTextContains(c.text, {"term": "microsoft", "distance":2})

Full-Text Search: New Languages

We’ve expanded language support for full-text search to include Italian (it-IT), Portuguese (pt-PT), and Brazilian Portuguese (pt-PR), joining French (fr-FR), German (de-DE), and Spanish (es-ES) preview languages. This makes it easier to build global, multilingual search experiences with built-in linguistic analysis and tokenization. Learn more.

You can define a full-text policy specifying multiple language analyzers as shown below.

Example:

{
   "defaultLanguage": "en-US",
   "fullTextPaths": [
       {
           "path": "/text1",
           "language": "it-IT"
       },
       {
           "path": "/text2",
           "language": "pt-PT"
       }
   ]
}

Full-Text Search: Faster Inserts & BM25 Scoring

Azure Cosmos DB indexing and query engine improvements to indexing and phrase search lower RU consumption and improve query responsiveness. The default BM25 scoring model has also been fine-tuned to improve relevance for longer or multi-term queries, especially beneficial for document retrieval, search, and e-discovery workloads

Azure Cosmos DB MCP Toolkit

The new MCP Toolkit helps developers integrate Azure Cosmos DB data directly into AI agent workflows, including Azure AI Foundry Agents Service. It provides an out-of-the-box way to retrieve structured, unstructured, and vector data from Azure Cosmos DB, simplifying the creation of RAG and memory-augmented systems. Learn more.

Private Preview: Semantic Reranking

Semantic Reranking (in private preview) can now be applied to any query type (including vector, full-text, or hybrid searches,) directly within Azure Cosmos DB. It uses Azure AI Search’s proprietary reranking model; the same model that powers its semantic ranker feature, to reorder results based on semantic meaning rather than raw keyword or vector similarity.

Developers can try it today via the .NET (Preview) and Python SDKs, enabling seamless integration into existing retrieval pipelines. This helps surface the most contextually relevant results for chatbots, Q&A systems, and AI agents. Sign-up for the private preview of Semantic Reranker.

Conclusion

The new search capabilities in Azure Cosmos DB make it easier than ever for developers to build intelligent, high-performance applications that combine structured, unstructured, and vector data. From fuzzy and multilingual full-text search to Float16 vectors and semantic reranking, these features bring richer, more efficient retrieval directly into your database. Whether you’re scaling a RAG pipeline or enhancing search in your app, Azure Cosmos DB gives you the flexibility and performance to build confidently for production at global scale.

About Azure Cosmos DB

Azure Cosmos DB is a fully managed and serverless NoSQL and vector database for modern app development, including AI applications. With its SLA-backed speed and availability as well as instant dynamic scalability, it is ideal for real-time NoSQL and MongoDB applications that require high performance and distributed computing over massive volumes of NoSQL and vector data.

To stay in the loop on Azure Cosmos DB updates, follow us on X, YouTube, and LinkedIn.