Microsoft 365 Copilot has launched Critique, a multi-model AI research tool that pairs GPT and Claude for more reliable outputs. This feature enhances research accuracy in enterprise workflows by using competing models to cross-check results.
How Critique Works
GPT from OpenAI generates the initial draft response to a research query. Claude from Anthropic then reviews it for factual accuracy, completeness, and proper citations before final delivery to users. Microsoft plans to enable bidirectional workflows, where Claude drafts and GPT critiques, creating a checks-and-balances system.
Early tests show a 13.8% gain on the DRACO deep-research benchmark, surpassing single-model tools from OpenAI, Google, Perplexity, and Anthropic. It reduces AI hallucinations, a top enterprise concern, by mimicking peer review processes.
Companion Features
Model Council displays side-by-side responses from multiple models for user comparison. Copilot Cowork, powered by Claude tech, handles multi-step tasks across Microsoft 365 apps and is available via the Frontier early access program.
With 15 million paid seats (3.3% of 450 million subscribers) as of Q2 FY26, Microsoft aims to accelerate enterprise adoption by improving reliability. This multi-model push evolves its OpenAI partnership while integrating Anthropic’s Claude.
0 Comments