OpenSearch 3.5 Is Now Generally Available on the NetApp Instaclustr Platform

OpenSearch 3.5 is now available on the NetApp Instaclustr Managed Platform — bringing major upgrades for AI-powered search, vector performance, observability, and search relevance tuning. 

Here’s what’s new for OpenSearch on the platform. 

Agentic AI: Conversation Memory & Context Management 

AI agents built on OpenSearch can now maintain persistent conversation memory, capturing context, tool reasoning, and responses across multi-turn interactions. New hook-based context managers dynamically optimise what gets sent to your LLMs — supporting truncation, summarisation, and sliding window strategies to reduce token costs and prevent context overflow. 

58% Faster Vector Search 

OpenSearch 3.5 introduces significant improvements to k‑NN vector search, delivering up to 58% higher throughput compared to the previous FP16 processing implementation. This gain comes from a new SIMD optimised execution path, which improves efficiency in multi‑segment scenarios while maintaining high recall and result quality. Additional improvements include more efficient filters, better nested query handling, and enhanced ef_search tuning. 

More Flexible LLM Connectors 

ML Commons connectors now support custom-named actions and additional HTTP methods (PUT/DELETE), enabling full CRUD operations through a single connector — making it simpler to integrate with external LLM providers and REST APIs. 

Search Relevance Workbench — Now with LLM-as-Judge 

The Search Relevance plugin adds LLM-powered evaluation with customisable prompts, scheduled experiments for tracking quality trends over time, and a streamlined side-by-side comparison UI. Scale your relevance testing beyond manual judgments. 

Prometheus in Discover & New PPL Functions 

Query and visualise Prometheus metric data directly in OpenSearch Dashboards alongside logs and traces — with PromQL autocomplete and gauge metric support. The SQL plugin adds new PPL commands including streamstats, mvcombine, mvzip, and addtotals for richer data transformation. 

Query Insights Improvements 

Every top-N query now automatically captures username and roles for ownership visibility. Query source storage has also been optimised to reduce storage overhead by up to 58% compared to the default _source storage model, by minimising duplication and reconstructing data dynamically when needed. 

Lifecycle Guidance and Upgrade Recommendations 

With the release of OpenSearch 3.5, we have reviewed an updated the lifecycle states of older OpenSearch versions in accordance with our lifecycle policy. To ensure continued access to full support and SLA’s, we recommend that customers run GA versions of OpenSearch wherever possible.  

Getting Started 

OpenSearch 3.5 is available now for new and existing clusters via the console, API, or Terraform. To unlock AI Search features, enable the AI Search Plugin — see our full plugin list for details. Existing customers can schedule an upgrade by contacting our support team.