Added support for Grok 4.3, Claude Opus 5.5, Claude Sonnet 5.5, GPT-6, GPT-6.1 Sol, and Fable 5.1. Once an administrator enables them in Provider Configs, users can pick these models in conversations.

Grok models can now use the Web Search tool, so answers can include current information from the internet.

The provider setup wizard now suggests each provider's latest mid-tier model instead of older ones like GPT-4. New providers start with a current default that balances performance and cost.

Third-party packages were updated to fix known security vulnerabilities.

The sidebar toggle is now available only on mobile devices, providing a cleaner navigation experience across AI Portal.

When sensitive content is detected, users can also see a configurable message explaining how to request that an item be added to the DLP ignore list if they cannot override detections themselves.

Administrators can configure DLP ignore-list requirements, including optional reasons, usernames, and user-specific entries in addition to global entries. They can also control who may mark content as not sensitive and review those actions in the audit log. DLP detection can be overridden by everyone (default), by no one, or by users in the configured roles.

Administrators can add a custom disclaimer to a provider. The disclaimer is shown when a user selects that provider, helping communicate important usage requirements before accessing its models.

The underlying sensitive-data detection service has been updated to improve recognition accuracy and reduce false positives and false negatives.

New customer environments now include Backup Vault configuration, improving support for data protection and recovery operations.

Chat shortcuts and homepage shortcuts now provide a consistent experience, making it easier to discover and use the same actions from either location.

Improved sensitive-text detection for content separated by newlines, helping maintain more consistent protection across formatted or multi-line inputs.

Administrators can now investigate cost anomalies using a detailed cost dashboard with visibility into usage by user, model, and day. This granularity makes it easier to identify cost drivers and investigate unexpected changes in spending.

The composer now provides controls for creating and formatting numbered and bulleted lists. New list items are added automatically as you type, and pressing Enter on an empty list item exits the list.

Provider administration now displays the available subscription quota when the required permissions are present, making it easier to understand provider capacity.

DLP detections are now managed from their own administration page instead of the general system settings area. The page provides a clearer view of detected sensitive information, and several country-specific detection categories have been removed from the product.

AI Portal now supports Fable 5 and newer OpenAI model pricing, including separate handling for short- and long-context requests where applicable.

Several underlying dependencies were updated to address security vulnerabilities and improve protection against problematic pattern matching.

Composer's Prompt menu has a new option. The feature uses your role, industry, and any topics you provide to create between three and nine clear, structured prompts. These prompts follow our recommended prompt format, similar to those created using the Improve prompt feature.

The provider configuration table now shows both available and total subscription quotas, limits deployment selections to available capacity, and improves the readability of long provider details by truncating text and expanding it on hover. When switching providers, the table also defaults to the first available option.

Selecting files and folders from the file system is now smoother and more intuitive, making it easier to attach the right content to your conversations.

The Admin dashboard is now focused on the AI Adoption. Per user token usage is now available directly on the dashboard.

Newly created deployments now default to the Global Standard configuration, providing more consistent availability and performance out of the box.

Provider setup and management have been refined for a more reliable configuration experience.

Rate-capture job now automatically retries when the upstream service is temporarily rate limited, improving the reliability of usage and cost data.

Web search and document generation now provide clearer in-progress status details, making it easier to understand what the assistant is doing during longer-running requests. Model reasoning indicators have also been refined so internal status content does not appear in the final response.

A new guided wizard simplifies provider configuration and helps administrators complete setup more efficiently. The wizard also displays provider save errors instead of failing without clear feedback.

The favicon configuration experience now informs administrators that customers may need to perform a hard refresh before a newly configured favicon appears.

AI Portal now accounts for cached tokens when tracking model usage and normalizes model pricing to a per-million-token rate. Improved mapping between model variants and their billing models provides administrators with more accurate usage and spend calculations.

Files has been redesigned to give you a more organized and consistent view of your content.

Added support for Anthropic Opus 5 and OpenAI GPT-5.5.