NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules
Last release 2 months ago
04 Aug 2026
Release timing varies
gaps range from 9 days to 4 months
Nearly every release is documented
notes for 34 of 35 stable releases
1 version withdrawn
withdrawn after publishing
8 years old
508 releases · first in 2019
ML.DeleteJob: Added WithDeleteUserAnnotations. Should annotations added by the user be deleted.
ML.DeleteJob: Added WithDeleteUserAnnotations. Should annotations added by the user be deleted.ML.ResetJob: Added WithDeleteUserAnnotations. Should annotations added by the user be deleted.ML.StartTrainedModelDeployment: Added WithPriority. The deployment priority.TransformGetTransformStats: Added WithTimeout. Controls the time to wait for the stats.TransformStartTransform: Added WithFrom. Restricts the set of transformed entities to those changed after this time.New APIs
TransformScheduleNowTransform documentation.
HealthReport documentation.
Changes
Do method on endpoints now return a typed response, one per endpoint.Perform method added on endpoints, returns http.Response as did Do.types.ElasticsearchError with .As and .Is methods..Raw now takes a reader as input._source in Hits are now json.RawMessage to highlight they later deserializable nature.AdditionalProperties, like the ones found in multi-bucket aggregations, are not yet supported.
One column per quarter.
Nothing published for this version
Nothing published for this version
ML.StartTrainedModelDeployment: Added WithPriority
ML.StartTrainedModelDeployment: Added WithPriorityNew APIs
ML.UpdateTrainedModelDeployment: Updates certain properties of trained model deployment.BulkIndexer
Improvements were made to the BulkIndexer memory usage to allow better handling under burst use cases. Thanks to @christos68k and @rockdaboot !
Nothing published for this version
Nothing published for this version
ML.StartTrainedModelDeployment: Description of NumberOfAllocations has been changed in "The total number of allocations this model is assigned across
ML.StartTrainedModelDeployment: Description of NumberOfAllocations has been changed in "The total number of allocations this model is assigned across machine learning nodes".Security.GetAPIKey: Added WithLimitedBy boolean parameter. Flag to show the limited-by role descriptors of API Keys.Security.GetUser: Added WithProfileUID boolean parameter. Flag to retrieve profile uid (if exists) associated to the user.Security.GetUserProfile: Changed the description of uid parameter, a comma-separated list of unique identifier for user profiles.Security.QueryAPIKeys: Added WithLimitedBy boolean parameter. Flag to show the limited-by role descriptors of API Keys.TextStructureFindStructure: Added EcsCompatibility string parameter. Optional parameter to specify the compatibility mode with ECS Grok patterns - may be either 'v1' or 'disabled'.Promoted to stable
ML.InferTrainedModelML.PutTrainedModelDefinitionPartML.PutTrainedModelVocabularyML.StartTrainedModelDeploymentML.StopTrainedModelDeploymentSecurity.activateUserProfileSecurity.DisableUserProfileSecurity.EnableUserProfileSecurity.GetUserProfileSecurity.HasPrivilegesUserProfileSecurity.SuggestUserProfilesSecurity.UpdateUserProfileDataNew APIs
ML.ClearTrainedModelDeploymentCache documentation.Security.BulkUpdateAPIKeys documentation.Indices.Downsample (Experimental API) documentationFollowing multiple feedbacks we decided to remove the builder API for the type tree.
In its place, work has started to further simplify the type tree by removing redundant type aliases. The API also now comes with a helper package named some that allows to call for inline pointers on primitive types.
In addition, a bug was fixed preventing the use of wildcards in index names, and enums are now extensible by default.
The Typed API remains in alpha stage while its development continues.
Nothing published for this version
Nothing published for this version
get, mget and search added force_synthetic_source: Should this request force synthetic _source? Use this to test if the mapping supports synthetic _so
get, mget and search added force_synthetic_source: Should this request force synthetic _source? Use this to test if the mapping supports synthetic _source and to get a sense of the worst case performance. Fetches with this enabled will be slower the enabling synthetic source natively in the index.ML.StartTrainedModelDeployment added cache_size: A byte-size value for configuring the inference cache size. For example, 20mb.Snapshot.Get added sort, size, order, from_sort_value, after, offset and slm_policy_filter. More on these in the documentation.New API
Security.UpdateAPIKey documentation.As highlighted in the release not for the 8.4.0-alpha.1, this release marks the beginning of the newly arrived TypedClient.
This new API is still in alpha stage and will be release alongside the existing esapi.
A few examples of standard use-cases can be found in the TypedAPI section of the documentation.
This second prerelease of the 8.4.0 updates the API for the client and fixes the serialization for types using additional properties.
This second prerelease of the 8.4.0 updates the API for the client and fixes the serialization for types using additional properties.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →