We’re continuing to improve the intelligence behind Microsoft 365 Copilot Declarative Agents. As part of this effort, Declarative Agents are being updated to use GPT‑5.2, a newer large language model designed to deliver higher‑quality responses across complex agent scenarios.
This update focuses on improving how agents reason, plan, and produce structured outputs - helping them perform better on the tasks that matter most to your organization.
Note: This update is gradually rolling out to all the tenants worldwide. Rollout is expected to be completed by end of March 2026.
What’s new with GPT‑5.2?
GPT‑5.2 brings meaningful quality improvements for agent experiences that involve:
- Reasoning and multi‑step tasks
Better handling of workflows that require planning, iteration, or decision‑making across multiple steps. - Tool usage and orchestration
More reliable tool calling and coordination when agents interact with data sources, APIs, or enterprise tools. - Document analysis
Improved comprehension and synthesis of longer or more complex documents. - Structured, well‑formatted outputs
Clearer tables, summaries, and formatted responses that are easier to consume and act on.
What does this mean for customers?
Because the underlying model is being upgraded, you may notice:
- Improved response quality across many agent scenarios
- Subtle changes in behavior for some prompts or workflows, especially those that rely on reasoning, formatting, or tool usage
These changes are expected and reflect the model’s improved capabilities. In most cases, agents should perform the same or better than before.
How will this update roll out?
The GPT‑5.2 update is rolling out gradually. Not all users within a tenant will receive the update at the same time. This phased approach helps ensure a smooth transition and allows us to monitor quality closely as the update expands.
What should admins and agent builders do?
No configuration changes are required to receive this update. However, we recommend a few simple steps to ensure a smooth experience:
For admins
- Inform agent builders that a model update is rolling out so they can plan light validation.
For agent builders
- Validate your most important scenarios
Run a quick check on your top workflows (for example, your top 5–10 prompts or use cases). - Confirm expectations are still met
Pay attention to accuracy, tone, reasoning, and formatting. - Adjust agent instructions if needed
Small prompt or instruction refinements can help agents take full advantage of GPT‑5.2’s capabilities.
Share feedback to help us improve
Your feedback plays a critical role in improving Copilot experiences. If you notice issues or unexpected behavior:
- Use thumbs up / thumbs down on agent responses
- Include #GPT52 in your feedback description to help us route it quickly
This helps our teams identify patterns, address issues faster, and continue improving quality.
Looking ahead
GPT‑5.2 is another step forward in making Copilot Declarative Agents more capable, reliable, and helpful for real‑world work. We’ll continue to evolve the models behind Copilot while prioritizing quality, transparency, and customer trust.
Stay tuned for more updates as we continue to enhance Copilot experiences across Microsoft 365.
Microsoft