The August 2025 PolyAI Agent Studio release focuses on performance improvements, voice configuration flexibility, and more powerful call review tools, as well as introducing
OpenAI’s new model GPT-5 as an option for agents.Expand the items for details:
Home page updates
The Agent Studio homepage now loads faster and presents key metrics more cleanly. The displayed metrics can be customized to meet the needs of a particular project, so what you see will be unique to your deployment.What’s new:
Reduced page load time
Improved widget layout for better readability
More consistent alignment across metric cards and charts
AI call summaries
Agent Studio now displays an AI-generated summary at the top of the Conversation Review screen.What’s new:
New AI Summary module in the Conversation Review screen
Generated post‑call using GPT summarization logic
This will only appear by default for projects that have signed the current DPA and support Azure AI..
Multi-voice limit increase
Projects can now have up to 10 active voices, updated from the previous limit of 5 voices.Voice limits apply per project and can be managed from the settings page.
New call review filters
It is now much easier to filter calls in Conversation Review by delivery status, channel, and other metadata.What’s new:
Faster filter response time
Filter combinations can be applied without resetting the view
Use saved filters to make these configurations instantly reusable.
Updated channel display
Conversation Review now displays additional call attributes and improved channel labels for quicker scanning.What’s new:
Clearer channel tags for Inbound, Outbound, and Agent Chat
Additional metadata columns available for custom table views
Streamlined column naming for consistency across the platform
GPT‑5 models in Agent Studio (experimental)
We’ve added GPT‑5 options to Agent Studio for evaluation and feedback.What’s available:
GPT‑5 nano and GPT‑5 mini: Recommended for most Agent Studio use cases (best balance of quality/latency)
GPT‑5 (large): Included for testing; note it’s currently slow at 3–4 seconds or more.
GPT‑5 chat (router): Routes dynamically across the above models so you can experiment quickly.
Switch models from function settings or your project’s model configuration panel.