What's new in Copilot Studio - Microsoft Copilot Studio (original) (raw)

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 roll 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.

May 2026

(General availability) Computer use is now generally available, letting your agents automate web and desktop apps by controlling browsers and desktop applications on behalf of users.
Add a prompt node to an agent flow or workflow to make a single AI call with dynamic content and model selection, useful for scenarios like translation and structured data extraction.
Add a Microsoft 365 Copilot node to a workflow to send prompts to Microsoft 365 Copilot or a specific agent, enabling automation scenarios like research and audit drafting.
Configure consent-based recording on voice-enabled agents to ask callers for consent before recording, with configurable compliance behavior and retention settings.
Use the agent inventory schema to discover and audit all Copilot Studio agents in your organization from the admin center, API, or Azure Resource Graph.
(Preview) Review agent readiness and issue status from a consolidated status page that surfaces publishing errors, runtime issues, and configuration blocks with severity levels.
Use asynchronous responses for agent flows to let long-running processes exceed the two-minute limit and return results to the agent when they complete.
(Preview) Automatically create Microsoft Entra agent identities for each of your agents to scope connector permissions, Conditional Access policies, and DLP governance to individual agents.
(Preview) Add computer use standalone tools to agents and agent flows for modular, reusable UI automation with built-in governance, model selection, and observability.
(Preview) Choose Mistral Medium 3.5 as the primary AI model for your agent, now available as an experimental option alongside Anthropic, xAI, and other supported providers.

April 2026

Configure hold and resume for voice-enabled agents to pause mid-conversation and resume where you left off, giving users a more natural calling experience.
(Preview) Build and deploy real-time voice agents for Copilot Studio and Dynamics 365 Contact Center, with support for NLU, multilingual conversations, knowledge integration, security settings, and voice tuning settings like speech sensitivity and silence detection. Configure and test your real-time voice agent directly in Copilot Studio, then publish it to your telephony channel.
Trigger agent evaluations automatically using the Copilot Studio connector with Power Automate, reducing manual effort in your testing workflow.
(Preview) Run automated agent evaluations from a REST API using the Power Platform API to integrate evaluation into CI/CD and custom automation pipelines.
Use the agent usage estimator to forecast Copilot credit consumption across Copilot Studio and Dynamics 365 agents before deploying at scale.
(Preview) Analyze your agent with custom metrics by defining your own analytics categories and visualizing results alongside built-in analytics.
(General availability) Connect agents to other agents using the agent-to-agent (A2A) protocol.
(Preview) Add a display name suffix to identify agents across environments in Teams and Microsoft 365 Copilot using environment variables.
(Preview) Use GPT-5.5 Reasoning (Deep) as an experimental model for agents requiring deep analytical reasoning.
Share analytics access with colleagues using the new Analytics Viewer role, without granting broader maker permissions.

March 2026

(Preview) Use Work IQ tools to connect Microsoft 365 Copilot and your agents to the Work IQ service. By using this connection, you can access real-time work insights and context from Microsoft 365 files, emails, meetings, chats, and more.
View and filter detailed lists of user questions and reactions from your agent's conversations. Use this data to identify gaps, create evaluation test sets, and export data for further analysis.
For users with Bot transcript viewer privileges, downloaded session transcripts from Copilot Studio now include session-level customer satisfaction (CSAT) values.
For users with Bot transcript viewer privileges, after you filter your list of questions or reactions, you can download the filtered questions or reactions to a .csv file.
Use the Prompt assistant in Prompt builder to draft prompts faster with GPT model–powered suggestions. This feature reduces the time needed to craft effective prompts.
Add Bing Custom Search as a knowledge source to ground your agent's responses in a curated, scoped web index by using a Custom Configuration ID.
Configure post-call action topics for voice-enabled agents to trigger backend actions automatically after a call ends, based on how the conversation concluded.
Configure reprompt messages in Question nodes with randomization support to improve conversation recovery for retry prompts, entity recognition failures, and voice scenarios.
Apply accessibility best practices for Adaptive Cards to support screen readers, keyboard navigation, and Teams-specific scenarios. These practices help you build more inclusive agent experiences.
Agent evaluations are now generally available (GA). You can validate agent performance using customizable test sets to help identify strengths and areas for improvement across diverse scenarios.
Create multi‑turn conversation tests in Copilot Studio to evaluate how agents perform across realistic dialog flows instead of single‑turn interactions.
ChatGPT‑5 is now generally available globally (excluding GCC environments) for use in production agents.
Claude Sonnet 4.5, Claude Sonnet 4.6, and Claude Opus are now generally available globally (excluding GCC environments). Use these models to optimize reasoning depth, quality, latency, and cost for your agents. Availability of some models might require data processing and storage outside of your organization's geographical boundaries. Learn more in Choose an external model as the primary AI model.
Use an agent node to call a Copilot Studio agent from an agent flow.

February 2026

Improved agent responses for ticket‑based Microsoft 365 Graph connectors. Agents more accurately retrieve ServiceNow tickets and Azure DevOps work items and generate clear, actionable summaries, which improves workflow reliability, and time to value.
Select Claude Sonnet 4.5 (beta) for Computer Use agents. This model improves nuanced decision‑making for complex tasks, increasing reliability, and success rates for advanced uses.
Enhancements to the prompt builder include:
- Configure content moderation sensitivity per prompt to control how hate/fairness, sexual, violence, and self‑harm content is filtered—supporting regulated and document‑processing scenarios with low or high sensitivity settings for managed models.
- Optimize prompts with new Claude models by choosing Claude Opus 4.6 or Claude Sonnet 4.5, enabling fine‑grained control over reasoning depth, quality, latency, and cost per prompt.
- Edit prompt instructions and settings inline in agent tool details, bringing model selection, inputs, knowledge, and testing into a single, streamlined authoring experience.

January 2026

(Preview) New enhancements to agent evaluations:
- Provide real‑time thumbs‑up/down feedback on evaluation results to verify grading performance and drive ongoing improvements to evaluation reliability.
- View your agent's sequence of inputs, decisions, and outputs with activity maps so you can quickly diagnose issues and get clearer insight into how your agent behaves at runtime.
- Use a validated CSV template to create test sets, reducing formatting errors, and helping your team standardize evaluation data more quickly.
(Preview) Expand computer use capabilities with new model support, built‑in credentials, enhanced audit logging with session replay, and Cloud PC pooling—giving you greater security, scalability, and governance for agent‑driven workflows.
(General availability) Use the Microsoft Copilot Studio extension for Microsoft Visual Studio Code to build, edit, and manage agents inside Microsoft Visual Studio Code, supporting advanced and highly flexible developer workflows.

December 2025

Compare multiple agent versions side by side to validate improvements and quickly spot regressions when evaluating agents with test sets.

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 Work IQ. Updated system architecture and new retrieval methods deliver more precise, context-rich responses, enhancing answer quality.