Release notes for Red Hat AI Inference

Use the release notes to learn about the latest changes to the documentation that are grouped by month.

Looking for IBM Cloud status, platform announcements, security bulletins, or maintenance notifications? See IBM Cloud status.

October 2026

Private endpoints with virtual private endpoint gateways: You can now connect to Red Hat AI Inference from your VPC over the IBM Cloud private network by using a virtual private endpoint (VPE) gateway. The private endpoint is available in the Washington DC (us-east) region and uses the same API and authentication as the public endpoint. For more information, see Using virtual private endpoints for VPC to privately connect to Red Hat AI Inference.
Create embeddings with Red Hat AI Inference: You can now use Red Hat AI Inference to generate text embeddings. Embeddings convert text into numerical vectors that capture semantic meaning, enabling semantic search, similarity comparisons, and retrieval-augmented generation (RAG) workflows. For more information, see What is embedding? and Improving chat completions with vector embeddings and RAG.

Model alignment is deprecated: Red Hat AI Inference model alignment, synthetic data generation, and taxonomy management features are deprecated and will be removed on 25 September 2026. To continue model customization and alignment workflows, migrate to Red Hat® OpenShift® AI On OpenShift. Learn more about Red Hat® OpenShift® AI.

New models available: Two new models are now available for inference: gemma-4-31b-it and granite-4-1-30b. To view all available models, open your project in Red Hat AI Inference and click Model catalog.

Inference is now generally available: You can now use Red Hat AI Inference to run inference workloads in production. Use industry-standard OpenAI compatible APIs for chat completions to integrate inference capabilities into your applications. A playground is now available in the console where you can test different inference settings and configurations before moving to production. For more information, see Getting started.

Inference with Red Hat AI on IBM Cloud (Beta): You can now use inference to interact with foundation models and evaluate AI-powered responses for your applications. The inference feature provides industry-standard OpenAI compatible APIs for chat completions and model management. This beta feature is available for evaluation and testing purposes. To get access to the beta, send an email to instructlab@ibm.com. For more information, see Working with chat completions.

Red Hat AI Inference CLI plug-in version 0.0.26: Version 0.0.26 of the plug-in adds file size limits. Each of the skills and knowledge documents must be less than 100 GB in size. The cumulative size of all the skills and knowledge JSON files must be less than 400 GB.

Version 1.5 of Red Hat AI Inference is available: For the best results, run training on newly generated synthetic data with version 1.5.; For more information, see the release notes, the RHEL AI documentation and the known issues
New base model: Red Hat AI Inference now uses the granite-3.1-8b-starter-v2.1 model. For more information, see the model specifications.

New! Import your own training data: You can now import your own training data for training models. When you import your own training data, you can specify previously generated data IDs, or add knowledge and skills files to a data generation job by uploading files from Object Storage or your local machine.
New! Taxonomy validation: When you upload a taxonomy to Red Hat AI Inference, it's now checked for formatting and syntax errors. Also, if you reference external knowledge documents in your qna.yaml files, Red Hat AI Inference checks for access to those files. Additionally, Red Hat AI Inference checks for the proper service authorizations for services like Object Storage and Secrets Manager.
Red Hat AI Inference CLI plug-in version 0.0.24: Version 0.0.24 of the plug-in adds support for importing your own training data to the data generate command.

New! Private repo support: You can now use Secrets Manager to give Red Hat AI Inference access to your taxonomy knowledge documents in private repositories or GitHub Enterprise repositories. You can enable private repository access when uploading your taxonomy in the console or by using the CLI. For information, see Getting started with Red Hat AI Inference.

Introducing Red Hat AI Inference on IBM Cloud!: Get ready to dive into AI! InstructLab is an open source project from IBM and Red Hat to be a cost-effective entry point into the world of machine learning.