Microsoft Puts Two AI Models to Work in Copilot's Research Tool

In Brief

Critique splits research between generation and expert review models.
System shows higher accuracy, depth and citation standards on benchmarks.
Council gives two models the same task then judges the results with a third model.

Are two AI models better than one? Microsoft thinks so and it has the DRACO scores to prove it. The company introduced Critique and Council in Researcher, the deep research agent within Microsoft 365 Copilot, on March 30. Both features integrate models from Anthropic and OpenAI, a shift from the previous single-model research workflows.

The difference between Critique and Council in a nutshell: Critique separates generation from evaluation, where one model handles planning, retrieval and drafting a report while the second acts as a reviewer. Council pits Claude and OpenAI against each other, asking each to create a report in parallel. It then brings in a third "judge" model to summarize where outputs agree, diverge or offer unique insights.

Researcher With Critique and Council Breakdown

Researcher with Critique

Critique's review process follows a structured format that focuses primarily on three elements:

Source Reliability
Report Completeness
Evidence Grounding

Researcher with Critique showed a 13.88% improvement over the next-highest system, Perplexity Deep Research (Opus 4.6), in DRACO benchmarking. It also showed a notable improvement over Researcher used without Critique, namely a 9.5 score improvement across four categories.

Microsoft Researcher With Critique scores vs. without

Council Model

Council is an alternate approach to Critique. Critique mode is the default when auto picker is set. Users can choose Council when they want to compare two competing outputs. The judge model produces an analysis of where and how the two outputs differed and flags similarities and unique points.

Both features are available through Microsoft's Frontier program.

The product announcement comes as Microsoft goes through a significant restructuring aimed at building AI momentum.

Microsoft Restructures for 'Workforce Acceleration'

Microsoft unified Copilot's consumer and commercial operations into a single organization across four pillars: Copilot experience, Copilot platform, Microsoft 365 apps and AI models in March. The reorganization saw Jacob Andreou take control of the product experience and Mustafa Suleyman refocused entirely on frontier model development. The news arrived against a backdrop of mounting pressure on its core AI product. Microsoft's January 2026 earnings call was the first time the company disclosed paid M365 Copilot adoption figures after eight quarters of silence: 15 million seats, representing 3.3% of its 450 million commercial Microsoft 365 base, according to analyst Jukka Niiranen.

Soon after the Copilot announcement, a second restructuring was announced, this time within the company's HR function. Chief People Officer Amy Coleman announced a full restructuring, bringing people analytics under employee experience, folding Culture and Inclusion under the new People and Culture team, eliminating the Chief Diversity Officer role and creating a new "Workforce Acceleration" team, tasked with skilling, redeployment and human-agent collaboration. Time will tell if the company will be able to continue its aggressive pace of product releases while undertaking a significant restructuring.

Have a tip to share with our editorial team? Drop us a line: