Review what's new and changed, and the fixes and limitations in AI Agent Studio for the v.40 release.

What's new

Expanded Model Support for Model connections

We have expanded our supported AI models for Model connections to provide enhanced reasoning and more robust generative automation capabilities. Automation developers and administrators can now select the following Standard models when creating connections:

Amazon Bedrock & Google Vertex AI

The full Anthropic Claude 4.5 suite is now available for both Amazon Bedrock and Google Vertex AI:

  • Anthropic Claude Sonnet 4.5
  • Anthropic Claude Haiku 4.5
  • Anthropic Claude Opus 4.5

Automation Anywhere

  • Anthropic Claude Sonnet 4.5
Deprecations:

Claude Sonnet 4:The Claude Sonnet 4 model has been deprecated and is no longer available for selection in the Model connection interface.

Create AI Skills

AI Agent supports additional Anthropic Claude 4.5 (Sonnet, Haiku, Opus) models

AI Agents now supports Amazon Bedrock and Google Vertex AI model connections that use Anthropic Claude Sonnet 4.5, Anthropic Claude Haiku 4.5, and Anthropic Claude Opus 4.5 (Standard type) models, as well as Automation Anywhere model connections that use Claude Sonnet 4.5 (Standard type). This enhancement allows AI Agents to use a broader range of large language models for planning, reasoning, and task execution beyond GPT-4o and GPT-4.1.

Create and manage AI Agent

What's changed

Enhanced package compatibility validation for AWS latest-generation models in AI Skill

For automation developers using the AI Skill command package with AWS latest-generation models (such as Haiku and Opus 4.5), the system now validates the minimum required package version before allowing execution.

When you select a Model connection that requires a newer AI Skill package version, the system displays an Incompatible package version message and blocks execution until you upgrade the package. The message clearly indicates the minimum required version.

In addition, if you downgrade the AI Skill package after configuring a compatible model connection, the system clears the incompatible selection and displays a warning.

This enhancement prevents runtime execution failures caused by package incompatibility and provides clear guidance to upgrade the package before using newer models. The minimum required version is dynamically validated using backend configuration to support future model updates.

Fixes

You can now close the citations modal window by clicking the Cancel button when using models that return citations, such as AWS Grounded by knowledge base. Previously, clicking Cancel on the citations modal window did not close it, requiring users to refresh the browser to dismiss it.

Limitations

Model connection errors due to package version mismatch or downgrade

In certain conditions you can encounter errors when connecting or testing a Model Connection (MC) if the package versions used in your environment are not aligned. This issue can occur in the following cases:

  • Version mismatch: The environment is configured with the default package version 10.x, but a bot or opportunity still references package 8.x.
  • Model Connection downgrade: A user creates an MC using the latest package version and then reverts the environment to an earlier version.
When this occurs, you might see an error such as: “Error while communicating to LLM provider. URL: — Error message: Agent Planner only supported for [Automation Anywhere, Azure OpenAI, OpenAI].”