What's new in 2026
Read about the latest features and capabilities of IBM® Content Assistant.
30 July 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.14 release in July 2026:
- Switch between LLMs without rewriting prompt templates
- You can now switch between different large language models (LLMs) without modifying your prompt
templates. Content Assistant uses
LLM-independent prompt templates that work with any LLM you choose, whether you use the default
prompt templates or customize them. Note: This feature is now available to all existing tenants who have not customized their prompt templates. Tenants who have customized their prompt templates will be automatically converted to the new LLM-independent format in a future release.
Learn more...
1 July 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.13 release in July 2026:
- Improve search accuracy with the new default embedding model
- You can now benefit from improved multilingual search capabilities with the new default
intfloat/multilingual-e5-largeembedding model. This model replaces the deprecatedibm/slate-125m-english-rtrvr-v2model and provides better search accuracy across multiple languages.Important: This change affects new indexes only. If you have existing indexes created with theibm/slate-125m-english-rtrvr-v2model, you must re-create them usingintfloat/multilingual-e5-largebefore August 8, 2026. After this date, the deprecated model will no longer be available on watsonx.ai and existing indexes using this model will stop working.
Learn more... - Switch between LLMs without rewriting prompt templates
- You can now switch between different large language models (LLMs) without modifying your prompt
templates. Content Assistant uses
LLM-independent prompt templates that work with any LLM you choose, whether you use the default
prompt templates or customize them. Note: This feature is available only to new tenants in this release. It will be available to all tenants in a future release. When rolled out, any customized prompt templates will be automatically converted to the new LLM-independent format, while default prompt templates will already use the new format.
Learn more...
- Block jailbreak attempts automatically with Granite Guardian
- Granite Guardian input moderation now automatically protects your enterprise from adversarial prompts that could cause policy violations, data leaks, or reputational damage. Enabled by default, the Granite Guardian LLM processes all questions from end users to perform input moderation before submitting them to the target LLM. When the system detects a jailbreak attempt, it rejects the request and displays an error message, maintaining your organization's security posture.
2 June 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.12 release in June 2026:
- Select a query mode for multi-document questions
- You can now select from three query modes when you ask questions about multiple documents in Content Assistant. Choose Precision Mode for cited answers from indexed documents, Full Context Mode (the default) for comprehensive responses, or Extended Precision Mode for maximum accuracy with citations.
Learn more... - Create watsonx ad hoc summaries from unindexed documents
- You can now generate watsonx ad hoc summaries for any documents in your selection without the
need to index them first. This feature eliminates the wait for the indexing process to complete and
provides immediate insights from unindexed or newly uploaded documents in your object store.
Learn
more... - Get document citations with full context in
GenAIMultiDocumentQueryAPI calls - The
GenAIMultiDocumentQueryAPI now supports a fourth processing mode called adaptive RAG mode. This mode sends both the most relevant document chunks and the complete document text to the LLM. Your applications can now deliver comprehensive answers with document citations based on both relevant chunks and complete document text.
Learn more... - Get longer AI responses with increased output token limit
- You can now receive AI responses up to 4096 tokens without truncation, a fourfold increase from
the previous 1024 token limit. This enables more comprehensive answers to complex queries, detailed
explanations, and complete document summaries without content being cut off.
Learn
more... - Enable adaptive RAG mode for the Ask Your Repository search
- As an administrator, you can now configure adaptive RAG mode for Ask Your Repository searches.
This mode sends both the most relevant document chunks and the complete document text to the LLM.
Your users can now receive comprehensive answers with document citations based on both relevant
chunks and complete document text.
Learn more... - Configure default chunk settings for search queries
- As an administrator, you can now configure how many document chunks Content Assistant sends to the LLM for
different types of searches. Set the default number of chunks (between 1 and 32) for single document
queries, multiple document queries, and repository-wide searches to optimize search behavior based
on your document types and the complexity of typical user queries. However, API callers can override
these defaults by specifying their own chunk values.
Learn more...
24 April 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.11 release in April 2026:
- Choose processing modes for
GenAIMultiDocumentQueryAPI calls - You can now optimize how the
GenAIMultiDocumentQueryAPI processes your multi-document queries by choosing from three flexible modes in Content Platform Engine 5.7 IF04 and later. Use RAG mode for selective chunk retrieval, entire document mode for full content, or adaptive mode for automatic selection based on document size.
Learn more... - Simplified IBM Content Navigator networking configuration for Content Assistant
- Starting with this release, IBM Content Assistant Client Managed
Software no longer uses the
IBM watsonx Assistant service. As a result, IBM Content Navigator and its web clients
no longer require network access to watsonx Assistant URLs to access Content Assistant features,
simplifying network and security setup.
Learn
more... - Expanded foundation model coverage and improved response quality
- Content Assistant
includes the following foundation model enhancements:
- Support for the
llama-4-maverick-17b-128e-instruct-fp8model. - Improved prompt templates for the
granite-4-h-smallmodel, enhancing response quality and accuracy.
Learn more...
- Support for the
11 March 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.9 release in March 2026:
- Improved single-document processing
- Single document questions now return complete and reliable answers. By default, Content Assistant now uses the Send entire document setting for single document queries, sending the full document instead of document chunks and giving the model complete context.
- Expanded model access through the default watsonx.ai account
- From this release, the default watsonx.ai account that is provided by the Content Assistant includes access to both IBM
provided models and open source foundation models. In earlier releases, access to open source models
required supplying your own watsonx.ai account.
Learn more...Important: If you run high volume production workloads, you need to continue to use your own watsonx.ai account. - Support for the OpenAI gpt-oss-120b model
- From this release, the Content Assistant supports the OpenAI
gpt-oss-120bmodel. This new model is currently available only in the IBM Cloud Dallas region. Learn more... - Model-aware relevancy score optimization
- This release introduces model-aware relevancy score optimization for embedding models. By using model-specific minimum and maximum relevancy score values, Content Assistant delivers more consistent and accurate results across different embedding models.
28 January 2026
The following feature updates were introduced in the IBM Content Assistant 2.0.8 release in January 2026:
- Content Assistant 2.0.8 package now available
- The Content Assistant 2.0.8
package for this release includes some updates to the
genai-ce.jarandgenaiPlugin.jarfiles.The Content Assistant can now index documents where the document titles contain either quotation marks or control characters.
Learn more...