Microsoft Introduces Critique in M365 Copilot for More Accurate Research

Microsoft launched Critique today, March 30, 2026, a new deep research multi-model system integrated into M365 Copilot. The announcement came directly from CEO Satya Nadella in a post on X, and the functionality is now available for participants in the Frontier program.

In practice, Copilot's Researcher now separates tasks. One model, usually based on OpenAI's GPT, handles planning, source searching, and creating the initial draft. Then, another model, most often Claude from Anthropic, acts as the critic: it fact-checks, improves citations, deepens analysis, and ensures greater objectivity. The end user receives only the refined version.

According to Microsoft's internal tests, Critique delivered +7 points in the DRACO benchmark, which measures accuracy, completeness, and objectivity in complex research. Compared to single-model tools from competitors like Perplexity, the gain reached 13.8%. This makes a difference, especially in long reports, market analyses, or documents that need reliable sources.

The feature runs in Auto mode by default, but there's also the Model Council, which displays responses from different models side by side to highlight agreements and divergences. Everything is integrated into the Microsoft 365 Graph, meaning Copilot can securely pull context from emails, files, and Teams conversations.

How This Changes the Game for Businesses

Instead of betting everything on having the "best model" at the moment, Microsoft is positioning itself as the orchestration layer. It combines the strengths of OpenAI and Anthropic without the user needing to switch between tools. Nicole Herskowitz, CVP of Microsoft 365, emphasized that the goal is to take the experience beyond simply using multiple models, making them work in a coordinated way.

Critique arrives alongside the expansion of Copilot Cowork in Frontier, another move that underscores the focus on agents that perform real tasks, not just answer questions. For Brazilian strategy, consulting, or marketing teams, this can significantly reduce the time spent on manual research and reviews, especially when the material needs to be robust for executive presentation.

Currently, access is limited to the Frontier program, which serves as an early testing channel for enterprise clients. Microsoft is expected to broaden availability in the coming weeks based on initial feedback.

Those who already use M365 Copilot daily should notice the difference first in heavy research tasks. The expectation is that this type of hybrid approach will become standard in corporate tools, as it reduces hallucinations and increases confidence without requiring the company to switch ecosystems.