Share via


What's new in Copilot Studio

This article provides resources to learn about new features in Copilot Studio.

Release plans

For information about new features being released over the next few months that you can use for planning, see Release Planner.

Released versions

For information about the new features, fixes, and improvements released in the past few weeks, see Released versions of Microsoft Copilot Studio.

Note

Releases are rolled out over several days. New or updated functionality might not appear immediately.

Notable changes

The following sections list features released in the past months, with links to related information.

November 2025

  • Updates for models used in Copilot Studio:

    • On November 24, 2025, GPT-5 Chat rolls out to Copilot Studio in general availability for European and United States regions. You can use generally available models for orchestration in production agents. For more information on choosing orchestration models, see Select a primary AI model for your agent.
  • (Preview) Automatically create Microsoft Entra agent identities for agents. When turned on, automatically apply identity management to individual agents by assigning a Microsoft Entra agent identity, helping admins secure and manage agents more effectively.

  • Improved knowledge retrieval for SharePoint-grounded agents using tenant graph grounding. Updated system architecture and new retrieval methods deliver more precise, context-rich responses, enhancing answer quality.

Note

Some queries might lead to slightly higher latency.

  • Improve response accuracy with SharePoint metadata filters. Use metadata like filename, owner, and modified date to refine knowledge retrieval and ensure responses come from the most relevant, up-to-date documents.

  • Orchestrate multiple agents to break down complex tasks across specialized agents, improving accuracy and speeding up end‑to‑end automation. Enhance your agents by linking them to other agents—either within your environment or external sources like Microsoft Fabric data agents—for modular, task-specific functionality.

  • (Preview) Add tool groups to agents for faster setup. Quickly equip your agents with curated sets of tools from Outlook and SharePoint connectors in one step. This streamlines setup, reduces errors, and ensures consistent, reliable orchestration.

  • Copy agents from Microsoft 365 Copilot to Copilot Studio. Easily move agents you created in Microsoft 365 Copilot into Copilot Studio to unlock advanced capabilities like multi-step workflows, custom integrations, and broader deployment options.

  • (Preview) Add human input to agent workflows with the request for information action. Pause an agent flow to collect details from designated reviewers via Outlook, then resume execution using their responses as dynamic parameters. This action ensures workflows can handle missing data or context without relying on hard-coded values.

  • Update Power Platform API calls to use the new 'copilotstudio' namespace. The previous namespace will continue to work temporarily, but switching now ensures compatibility with future updates.

  • Use component collections with new enhancements. Access collections directly from the sidebar, quickly export or import collections, and take advantage of support for primary agents and new connector types, including child agents and Model Context Protocol (MCP).

October 2025

  • Updates for models used in Copilot Studio:

    • Between October 27 and 31, 2025, GPT‑4o will be retired in Copilot Studio for agents using generative orchestration, except for GCC customers who can continue using GPT‑4o. The new default model is GPT‑4.1, which delivers improved performance, reliability, and consistency across experiences. GPT‑4o remains available until November 26, 2025 if you turn on the “Continue using retired models” option.
    • Choose from multiple AI models to tailor your agent's performance to your needs.
    • (Preview) Test and deploy GPT‑5 models to explore advanced capabilities and enhance your agent's performance.
  • Learn about Copilot Studio Kit, a suite of tools developed by the Power Customer Advisory Team (Power CAT) to help test custom agents, validate AI-generated content, analyze conversation key performance indicators, and more.

  • (Preview) Group related user questions into themes and drill down into analytics to uncover patterns and gain deeper insights.

  • Track time and cost savings for both autonomous and conversational agents to measure ROI and optimize performance.

  • Access a unified activity and transcript view, pin sessions, and submit feedback for faster, more effective troubleshooting.

  • (Preview) Accelerate flow execution to minimize timeouts and deliver a faster, smoother user experience.

  • Use the Model Context Protocol (MCP) server to access dynamic, real-time content—such as files, database records, and API responses—for richer context and improved agent responses.

  • (Preview) Evaluate your agents using customizable test sets—whether uploaded, manually created, or AI-generated. Test sets can include test cases using different test methods (graders) measured against defined reference answers, helping teams identify strengths and areas for improvement. This capability supports more reliable, high-quality agent experiences across diverse scenarios.

September 2025

  • (Preview) Automate tasks in desktop applications on Windows using Computer-Using Agents (CUA), which combines vision and reasoning to interact with interfaces—even when APIs aren’t available.

  • Embed Copilot agents into Android, iOS, and Windows apps using the Client SDK to provide rich, multimodal conversations within native experiences.

  • (Preview) Upload Excel, CSV, and PDF files for your agent to analyze using Python code, powered by the code interpreter in chat.

August 2025

  • (General availability) Use code interpreter to generate Python code-based actions from natural language in both the prompt builder and agent workflows.

  • (General availability) Enhance agentic response accuracy in Copilot Studio agents by using file groups to organize local files to be uploaded as a single knowledge source and apply variable-based instructions.

  • Allow users to upload files and images that your Copilot Studio agent can analyze and use to generate responses, then pass those files to downstream systems using Agent Flows, Power Automate, connectors, tools, and topics for seamless integration.

  • (General availability) Track and analyze unanswered queries and response generative AI quality with the generated answer rate and quality section in the Analytics page to improve your agent's performance.

  • Connect to an existing MCP server directly within Copilot Studio using a guided experience.

July 2025

  • Use advanced NLU customization to define topics and entities using your own data for higher accuracy and improved containment, especially for Dynamics 365 scenarios.

  • Search across your agent's knowledge, topics, tools, skills, and entities instantly using a new in-app search experience accessible via keyboard shortcut or top-level search.

  • Estimate time and cost savings based on successful runs or actions and customizable to your organization's metrics with ROI analytics for agents with autonomous capabilities.

  • View user comments submitted with thumbs up/down reactions in analytics, offering deeper insight into customer feedback on agent responses.

  • (Preview) Display Microsoft Information Protection (MIP) labels across connectors, test chat, Teams, and Microsoft 365 Copilot to prevent oversharing and support secure, compliant AI experiences. With new integrations between Copilot Studio, Dataverse, and Microsoft Purview, you can automatically classify sensitive data and ensure agents respect Purview sensitivity labels.

  • Publish agents directly to a WhatsApp phone number, making it easier to reach customers.

  • (Preview) Streamline authentication for Microsoft Entra ID–backed actions and connectors with the SSO Consent Card by allowing users to grant consent directly within the chat with no redirects and no interruptions.

June 2025

  • Improved experience for tools:

    • Grouping and filtering for easier search and discovery of tools.
    • Support for IntelliSense automatic completion and input widgets such as calendar control, file picker, and timezone picker, when configuring tools.
    • Improved tool invocation experience for customers with more affordances for complex inputs and clearer error messaging.
    • Automatic detection of SSO for connectors.
  • (Preview) Support for Microsoft 365 Copilot Tuning to train models on your own enterprise data for domain-specific tasks and integrate these models into Microsoft 365 experiences like Copilot in Teams, Word, and Chat. You can also connect your fine-tuned models to custom agents.

  • Actionable insights for questions that generative AI left unanswered, grouped by themes, in the Answer rate and quality section of the Analytics page.

  • Knowledge sources analysis for autonomous agents.

  • Ability to insert Power Fx formulas directly in the embedded prompt builder prompt editor.

  • Simplified text validation and extraction with regular expression support for Power Fx formulas that use IsMatch, Match, or MatchAll functions.

  • (Preview) Support for file groups as knowledge sources.

  • (Preview) Generative orchestration available for all supported languages.

  • Redesigned Channels page.

  • (US-only preview) Ability to select the GPT-4.1 mini experimental response model for generative answers.

May 2025

April 2025

  • (Preview) Support for Customer Managed Keys.

  • Support for generative and autonomous agents in Analytics.

  • Inclusion and visibility of agent data and ROI analysis in Viva Insights.

  • Nine new Graph connectors: Guru, GitLab (Issues, Merge Requests, Knowledge), Asana, 15Five, Miro, Trello, Zendesk Ticket, SmartSheet, and Seismic.

March 2025

February 2025

January 2025

  • Microsoft Purview can access and manage user interaction audit logs from Copilot Studio agents.

  • Chinese (Traditional) (zh-TW) language support for interactions between agents and users, including support for generative answers.

  • Default data policy enforcement for agents is set to Soft-enabled for all tenants.