Microsoft Corp. announced two new features for its Copilot Researcher tool on Monday. The update pairs OpenAI and Anthropic models to improve deep research capabilities for enterprise users. This move targets the growing competition in autonomous AI research tools across the technology sector.
The new modes, named Critique and Council, allow multiple models to work on the same task concurrently. Critics note that single-model systems often produce hallucinations or weak citations when handling complex queries. Combining models aims to fix these reliability issues through collaboration and verification steps.
Key Technical Details
Critique separates generation from evaluation using a sequential workflow designed for accuracy. One model drafts the report while a second reviews the content for factual precision and tone. Microsoft states this workflow reduces factual errors significantly by adding a layer of human-like verification.
Critique initially places GPT in the drafting role while Claude handles the review process. Engineers noted the system could eventually run in the opposite direction with Claude drafting and GPT critiquing. This flexibility allows organizations to test different model combinations for their specific needs.
Council runs GPT and Claude simultaneously to compare results side by side for the user. A third judge model summarizes agreements and divergences between the two systems automatically. This approach allows users to see different perspectives on the same query without manual comparison.
Microsoft tested the system on the DRACO benchmark covering one hundred complex tasks across ten domains. Copilot with Critique scored 57.4 points, surpassing Anthropic Claude Opus 4.6 by nearly 14% in the test. The improvement focused on analysis breadth, presentation quality, and factual accuracy in fields like medicine and law.
Market Implications
Major tech companies have been racing to build research agents this year to capture market share. Google and OpenAI released similar tools, but Microsoft chose a multi-model approach for its enterprise customers. This strategy leverages partnerships rather than relying on a single internal model for all tasks.
xAI and Perplexity also entered the market with their own research capabilities recently. These competitors focus on single-model efficiency while Microsoft prioritizes cross-model verification. Analysts suggest this could redefine standards for AI reliability and trust in business environments.
"Critique is a new multi model deep research system designed for complex research tasks," Microsoft explained in its official documentation. The company emphasized separating generation from evaluation to ensure quality standards are met. Official documentation outlined the specific roles each model plays in the process to reduce ambiguity.
Users need a Microsoft 365 Copilot license costing $30 per user per month for access. Access also requires enrollment in Microsoft's Frontier early-access program for these specific features. These features are currently available to enrolled members only until general rollout occurs.
The industry may shift toward hybrid model architectures for high-stakes tasks requiring high accuracy. Future updates may allow models to swap roles during the workflow dynamically based on task complexity. This development signals a maturation of agentic workflows in business software and enterprise tools.
Microsoft positions this as a significant step forward in enterprise AI research and safety. The integration highlights the value of cooperation between competing model providers in the open market. This development signals a maturation of agentic workflows in business software and enterprise tools.
