June 2026 release notes for ScreenPlay in Automation Cloud.

ScreenPlay coded workflows now handle two additional argument types: SecureString, for passing sensitive string values without exposing them in plain text, and ScreenPlayImage, for including a screenshot directly in the ScreenPlay prompt to give the model additional visual context. Previously, coded workflows had no native handling for these types; this change brings them to parity with low-code workflows.

ScreenPlay now adds an X-UIPATH-SEMANTIC-SPAN-ID header to the HTTP calls it makes to backend services, such as the Semantic Proxy and LLM Gateway. The span identifier lets a single ScreenPlay execution and its requests be correlated end-to-end across services, improving observability and troubleshooting.

ScreenPlay now supports the Application instance window attach mode. When a Use Application/Browser activity uses this attach mode, inner activities search across the entire application instance — all parent and child windows, alerts, and popups — instead of a single window.

ScreenPlay captures and records traces correctly across all windows of the attached application instance, in addition to the existing Single window support.

The ScreenPlay HTML trace file has a refreshed visual design. The trace viewer output now has an updated layout and styling that make execution traces easier to read and navigate, aligned with current UiPath design standards.

The following entries were omitted from the original June 24, 2026 announcement and apply to that release.

The Model drop-down on the ScreenPlay activity now offers a wider range of Screen Agent models, adding GPT-5.4, GPT-5.4 mini, GPT-5.5, Gemini 3.0 Flash Preview, and Path Mini. The added models are more capable and faster than the ones previously available, and are the largest contributor to the reduced execution latency in this release compared to the 25.10 GA release. Path Mini is a Qwen-based model fine-tuned by UiPath for screen understanding and UI interaction, hosted and served by UiPath on its own cloud infrastructure, and is the fastest option in the selection.

The full selection is:

Basic tier:

Standard tier:

For the characteristics of each model, refer to ScreenPlay.

The default model for the ScreenPlay activity changed from Screen Agent (using Gemini 2.5 Flash) to Screen Agent (using GPT-5.4). GPT-5.4 is a Standard tier model, whereas Gemini 2.5 Flash is a Basic tier one. Existing automations keep the model they were configured with; the new default applies to newly added ScreenPlay activities.

ScreenPlay is no longer confined to the application or window passed as its scope. It can now work across multiple browser tabs and multiple applications within the same task.

This release includes fixes to ScreenPlay execution reliability.