OpenSearch 3.5 and 3.6 supercharge AI workflows with smarter search and memory tools

Faster, leaner, and more precise: OpenSearch's new releases redefine AI performance. Developers gain tools for agentic systems, binary quantization, and real-time debugging.

, and Administrator

2026 May 2 . 11:06 PM

1 min read

The image shows a screenshot of a website with a number of items on it, including text, numbers,... — The image shows a screenshot of a website with a number of items on it, including text, numbers, and a search bar. It appears to be a Google search for "we are women".

OpenSearch 3.5 and 3.6 supercharge AI workflows with smarter search and memory tools

OpenSearch has released updates in versions 3.5 and 3.6, focusing on AI integration and performance improvements. The latest upgrades introduce new tools for agentic workflows, memory efficiency, and vector search capabilities. Developers now have access to features designed for faster, more scalable AI applications. OpenSearch 3.5 brought agentic conversation memory into the ML Commons, allowing better context management through hook-based systems. This version also laid the groundwork for Better Binary Quantization (BBQ), a method to compress high-dimensional float vectors into binary formats, cutting memory use by 32 times.

Version 3.6 expanded these capabilities with a multi-agent orchestration layer, supporting the Model Context Protocol (MCP). Agents can now search stored memory using vector similarity, keyword matching, or a mix of both. The update also fixed an async encryption issue, reducing thread contention and race conditions in high-throughput setups.

For search accuracy, BBQ recall on the Cohere-768-1M dataset reached 0.63 at 100 results, outperforming Faiss Binary Quantization’s 0.30. The release added BBQ flat index support for exact-recall tasks and introduced the SEISMIC algorithm for neural sparse approximate nearest neighbor search. Token usage tracking was also integrated, automatically logging counts per turn and model without extra configuration.

Debugging tools improved with Application Performance Monitoring, built on OpenTelemetry standards. This helps trace and analyse multi-step agent executions more effectively. The updates in OpenSearch 3.5 and 3.6 strengthen its role as a foundation for AI-driven applications. Enhanced memory handling, search precision, and monitoring tools provide developers with more control over complex workflows. These changes aim to support scalable, high-performance agentic systems in production environments.