What's new and changed in Watson Machine Learning

IBM Cloud Pak for Data Version 5.4.0

A new version of Watson Machine Learning was released in June 2026.

This release includes the following changes:

New features

This release of Watson Machine Learning includes the following features:

New job operator role for deployment spaces

You can now assign the job operator role to collaborators in deployment spaces. This new role is designed for production environments where you need to delegate job execution and scheduling responsibilities without granting full editing privileges.

The new job operator role is available when adding collaborators to deployment spaces. Job operators can manage job runs while maintaining the integrity of job definitions and assets. When a job operator runs a job, the job executes with the privileges of an editor or admin who has job permissions delegated to them in the deployment space.

By using the job operator role, you can achieve the following benefits:

• Enhanced security: Separate job execution responsibilities from asset modification privileges

• Production readiness: Enable operations teams to manage job runs without risking changes to production assets

• Flexible delegation: Allow editors and admins to delegate routine job management tasks while maintaining control over definitions

• Audit compliance: Maintain clear separation of duties for regulated environments

For more information, see Deployment space collaborator roles and permissions.

Scale your AI workloads with asynchronous execution

You can now run AI functions asynchronously to improve performance and scalability for I/O-bound tasks.

When you define async generator functions, the runtime automatically detects and executes them concurrently, enabling external API calls, file I/O, and parallel async tasks.

For more information, see Writing deployable Python functions.

Deploy ML models with the new onnxruntime_opset_21 software specification

You can now deploy machine learning models with the new onnxruntime_opset_21 software specification. The specification provides enhanced performance and compatibility with the latest ONNX model formats.

For more information, see Software specifications.

Configure runtime-specific ephemeral storage to ensure predictable scheduling and safer capacity planning

You can now configure ephemeral storage requests and limits for each runtime type in AutoAI. By setting these values in the WmlBase custom resource, you prevent runtime pods from being evicted when disk space runs low and ensure that workloads have the temporary storage they need for model downloads, scratch data, and logs.

New features from 5.3.1 patches

This release of Watson Machine Learning includes the following features that were introduced in IBM® Software Hub Version 5.3.1 patches:

Deploy machine learning models exported to ONNX 1.17

From release 5.3.1, patch 2, you can deploy machine learning models that were exported to the ONNX 1.17 format.

Updates

The following updates were introduced in this release:

Removal of software specifications that are based on Runtime 24.1

All software specifications that are based on Runtime 24.1 are now removed. If you have any assets that use the old software specifications, you must recreate them by using the latest corresponding software specifications.

For more information, see Software specifications.

Customer-reported issues fixed in this release

For a list of customer-reported issues that were fixed in this release, see the Fix List for IBM Cloud Pak for Data on the IBM Support website.

Deprecated features

The following features were deprecated in this release:

Deprecation of do_20.1 software specification for Decision Optimization

The do_20.1 software specification for Decision Optimization is deprecated and will be removed in a future release. If you have any assets that use the deprecated software specification, migrate them to the latest corresponding software specification.

For more information, see Software specifications.

The following features were deprecated in patches for release 5.3.1:

Software specifications that are based on Apache Spark 3.4 are deprecated

Software specifications that are based on Apache Spark 3.4 are deprecated since release 5.3.1, patch 2 and will be removed in a future release. To avoid disruption, migrate existing deployments that use Apache Spark 3.4 to Apache Spark 3.5.