Elasticsearch 8.11: Announcing GA of Elastic’s Learned Sparse EncodeR
Elasticsearch 8.11, published November 7, 2023, makes ELSER generally available for production semantic search and adds native Elastic Cloud connectors.
By Dillip Chowdary • Nov 16, 2026 • Source: Elastic Blog

At a glance
- Product
- Elasticsearch 8.11
- Model
- ELSER generally available
- Introduced
- ELSER first introduced in 8.8
- Published
- November 7, 2023
- Availability
- Elastic Cloud, Stack, ECE, ECK
- Vector limit
- Float arrays >128 and <4096
In this post
11 facts.The source has no prices, benchmark scores, or install flags, so the table and availability notes will use only the mapping limits and release facts it actually states. I'll count the draft against the 800–1,000 word band before sending it.The draft is about 110 words over the cap, and one table row states a “before” the post never gives.Elasticsearch 8.11 makes the Elastic Learned Sparse EncodeR model, ELSER, generally available for production semantic search, including integration with transformer models a team chooses. Elastic Blog's report, by Srikanth Manvi on November 7, 2023, also covers generative AI updates, simpler dense vector mappings, Search Labs additions, and sync from Outlook, Teams, Zoom, and Box. It is available now on Elastic Cloud, the only hosted Elasticsearch offering Elastic says includes every new feature, and as downloads of the Elastic Stack, Elastic Cloud Enterprise, and Elastic Cloud for Kubernetes.
This piece is for search engineers and platform teams running semantic search or internal knowledge search on Elastic Cloud or a self-managed cluster. It covers the two ELSER model profiles, LlamaIndex and LangChain vector-store support, indexing-buffer and dense_vector changes, Workplace Search maintenance, and the connector catalog with Extraction Service. Teams already ingesting through the web crawler or connectors will also find the UI default for the managed ELSER pipeline.
01Two ELSER profiles for production
ELSER is the retrieval model Elastic trained for semantic search. Results follow contextual meaning and user intent rather than exact keyword matches. Elastic introduced ELSER in 8.8, and 8.11 makes it generally available for production with improved relevance and performance.
Users now see two ELSER model IDs in 8.11, along with a recommendation on which one to use based on their profile. The Universal Version can run on any hardware. The Optimized Version is for the x86 family of architectures.
| Model profile | Hardware |
|---|---|
| Universal Version | Any hardware |
| Optimized Version | x86 family of architectures |
Teams already on ELSER can move to the latest version, and a notebook in Elasticsearch Labs walks through that upgrade. With the web crawler or connectors, the managed ingest pipeline defaults to the latest ELSER when the deployment is created in the UI. Elastic describes production ELSER as the route to best-in-class relevance.
02LlamaIndex, LangChain, and Labs
Elasticsearch can now be used as a vector store with LlamaIndex. Elastic says interest in Elasticsearch as a vector store continues in the LangChain community, and that Elasticsearch is the most upvoted vector store integration on LangChain. Both are part of the generative AI developer updates in this release.
Search Labs continues to add tutorials. "Use Amazon Bedrock with Elasticsearch and Langchain" covers splitting fictional workplace documents into passages, turning those passages into embeddings in Elasticsearch, and integrating an Amazon Bedrock LLM. "How to create customized connectors for Elasticsearch" covers custom connectors that simplify ingestion.
A third Labs entry, "Less merging and faster ingestion in Elasticsearch 8.11," states that Elasticsearch 8.11 improves how it manages its indexing buffer, resulting in less segment merging. Elastic names that piece among the notable recent Search Labs entries in this release.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
03Default dense vector indexing
The dense_vector field type supports indexing, and indexing is now enabled by default, so mappings do not have to set index:true. Elastic says this release makes Elasticsearch a great vector database through more intuitive, more performant dense vector mappings and added Search Labs content. With the defaults below, a dense vector mapping can be defined in one line.
| Setting | What changed | Result in 8.11 |
|---|---|---|
| index:true on dense_vector | No longer required in the mapping | Indexing is on by default |
| Dimension count | Need not be defined in advance | If undefined, set from the first ingested document with that field |
| similarity | Field may be left undefined | Defaults to cosine |
| Float array width | Greater than 128 and less than 4096 | Automatic dense_vector mapping |
If dimensions are left undefined, the count is set when the first document with a value for that field is ingested, rather than written into the mapping beforehand. An undefined similarity field defaults to cosine. Float arrays greater than 128 and less than 4096 are mapped automatically as a dense_vector.
04Workplace Search maintenance mode
Elastic is adding machine learning features on the Elasticsearch platform for emerging use cases, and it is changing Workplace Search at the same time. The standalone Workplace Search product will be in maintenance mode and will receive only security updates going forward. Elastic says that use case remains important for Elastic and for its user community.
Elasticsearch is the product Elastic recommends for internal knowledge search. The extended connector catalog gives customers who were evaluating Enterprise Search Workplace Search content sources a full path to sync that data. Those synced sources are meant to be used fully with ESRE capabilities.
05Native connectors and Extraction Service
Elastic says unified knowledge search can use content from private sources such as Outlook, Teams, Zoom, and Box. In 8.11, Google Drive, OneDrive, and GitHub are available as native connectors on Elastic Cloud. Customers on the Google Drive, OneDrive, and GitHub connector clients can switch to those native connectors and offload self-managed infrastructure cost.
Dropbox, ServiceNow, and GitHub are generally available in this release. OneDrive and Network Drive add document-level security so shared data is exposed to the right privileged audience. GitHub is listed both as a native Elastic Cloud connector and among the generally available connectors.
Ingest Pipelines extract binary content from small files only. Larger files, measured in gigabytes, can be handled by connector clients together with Extraction Service. Elastic says most of its connectors support binary content extraction on edge for those large files.
Note: Existing Elastic Cloud customers can reach many of these features from the Elastic Cloud console, and Elastic offers a free trial for teams not already on Elastic Cloud. Release timing for any feature described here stays at Elastic's sole discretion, and a feature that is not available now may not be delivered on time or at all.
Frequently asked questions
Is ELSER production-ready in Elasticsearch 8.11?
Yes. Elastic introduced ELSER in 8.8, and Elasticsearch 8.11 makes the Elastic Learned Sparse EncodeR generally available for production semantic search, with improved relevance and performance. Results follow contextual meaning and user intent rather than exact keyword matches.
What is the difference between the two ELSER model IDs?
Elastic shows two model IDs and recommends one based on the user's profile. The Universal Version can run on any hardware. The Optimized Version targets the x86 family of architectures.
What dense_vector defaults changed in 8.11?
Indexing is enabled by default, so mappings do not need index:true. An undefined dimension count is set from the first ingested document that contains the field, undefined similarity defaults to cosine, and float arrays greater than 128 and less than 4096 map automatically as dense_vector.
What is happening to Workplace Search?
The standalone Workplace Search product moves to maintenance mode and will receive only security updates. Elastic recommends Elasticsearch for internal knowledge search. Customers who were evaluating Enterprise Search Workplace Search sources get a connector-catalog path that uses those sources with ESRE.
How does Extraction Service handle large files?
Ingest Pipelines extract binary content from small files only. Files measured in gigabytes go through connector clients together with Extraction Service, and Elastic says most connectors support that binary extraction on edge.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement