Elasticsearch 8.3: What's New, Release History & Support Lifecycle

What Is New in Elasticsearch 8.3

Elasticsearch 8.3 brings significant enhancements across vector search, analytics, and data management. This release focuses on making complex operations more efficient and developer-friendly.

How does vector search get better in 8.3?

The headline feature is the introduction of approximate nearest neighbor (ANN) search. This is a game-changer for performance on large-scale vector similarity operations.

ANN search uses the HNSW algorithm to find similar vectors without an exhaustive, costly scan. In practice, this means your k-NN queries return results orders of magnitude faster, especially on big datasets. You enable it by setting method to hnsw in your field mapping.

PUT my-index
{
  "mappings": {
    "properties": {
      "my_vector": {
        "type": "dense_vector",
        "dims": 3,
        "index": true,
        "similarity": "l2_norm",
        "index_options": {
          "type": "hnsw",
          "m": 16,
          "ef_construction": 100
        }
      }
    }
  }
}

What analytics upgrades should I know about?

Elasticsearch 8.3 supercharges geospatial and terms-based analytics. The new vector tiles API delivers map visualizations significantly faster than the previous geo_shape approach.

This matters because rendering complex maps for dashboards used to be a bottleneck. Now, you can serve high-resolution map layers directly from Elasticsearch without overloading your Kibana instance. Under the hood, it uses the Mapbox Vector Tile specification.

For terms enrichment, the new terms policy type lets you enrich incoming documents based on exact keyword matches. It's simpler and more efficient than the match policy for use cases where you don't need fuzzy matching.

Are there any scripting enhancements?

Yes, Painless scripting now supports the long and double field types for the field API. This closes a previous gap where you had to work with these types differently.

Before, you had to use doc['my_long_field'].value. Now, you can use the cleaner and more intuitive field('my_long_field') and field('my_double_field') methods directly in your scripts. It makes writing scripts for numeric calculations much more straightforward.

// Now you can do this:
def my_calc = field('price').get(0) * field('tax_rate').get(0);

// Instead of this:
def my_calc = doc['price'].value * doc['tax_rate'].value;

How is manageability improved?

Two key areas see major improvements: the frozen tier and async search. Searches on the frozen tier are now up to three times faster, drastically cutting the cost and time needed to query archived, read-only data.

For async search, the protocol for communication between coordinating nodes and the node holding the async search results has been optimized. This reduces overhead and makes checking the status of long-running searches more efficient across the cluster.