Microsoft Pairs GPT and Claude for Collaborative Research in Copilot Researcher

Microsoft has announced two new capabilities for its Copilot Researcher tool — Critique and Council. Both features leverage GPT models from OpenAI and Claude from Anthropic simultaneously, directing them to work on the same task in tandem to deliver more accurate results.

"Introducing Critique, a new multi-model deep research system in M365 Copilot. You can use multiple models together to generate optimal responses and reports." — Satya Nadella (@satyanadella), original post

Why This Matters

According to Microsoft, every existing AI research tool operates the same way — a single model generates an answer with no independent verification. This leads to hallucinations, citation errors, and false or inaccurate claims. The multi-model approach represents a systematic attempt to address one of generative AI's most persistent problems — not by improving a single model, but by building a framework where models check each other's work.

Critique: Models That Collaborate

Critique is a multi-model deep research system designed specifically for complex investigative tasks. It combines neural networks from leading AI labs using a role-based division of labor.

As Microsoft explained, one model handles the generation phase — planning the task, iterating through information searches, and producing an initial draft. The second model acts as an expert reviewer, verifying and refining the output. Notably, the models can swap roles during the process.

DRACO benchmark results: Critique scored 57.4 points versus 42.7 for Claude Opus 4.6. Source: Microsoft

On the DRACO benchmark, which covers 100 complex research tasks across 10 domains, Critique achieved a score of 57.4 points. By comparison, Claude Opus 4.6 operating alone scored 42.7.

Council: Models That Compete

The second feature — Council — takes a fundamentally different approach. GPT and Claude run in parallel, each producing a full independent report. Both responses are displayed side by side, and then a third model serves as a judge, reading both outputs and writing a summary. This summary highlights where the two models agree and where their arguments diverge.

The core distinction between the two modes is straightforward: in Critique, models collaborate; in Council, they compete.

Availability

Critique is the default mode in Researcher, while Council must be manually selected from a menu. Both features are currently available to users enrolled in the Microsoft Frontier program.

Earlier in March, Microsoft expanded its AI ecosystem by introducing the Cowork agent for handling complex tasks within Microsoft 365. Cowork can interact with ecosystem applications including Outlook, Teams, Excel, PowerPoint, and SharePoint.