Microsoft's Copilot Critique Uses GPT and Claude Together - One Drafts, the Other Fact-Checks

Microsoft just launched Critique, a multi-model deep research system inside M365 Copilot. It runs GPT and Claude in sequence - GPT drafts the response, then Claude reviews it for accuracy, completeness, and citation quality before anything reaches the user. The idea is straightforward. One model generates, another evaluates. Microsoft is treating its AI partnerships as a supply chain rather than picking a single provider. It also ships a Council mode that runs models side by side for comparison, and reports a nearly fourteen-point improvement on the DRACO deep research benchmark. This is notable because Microsoft is openly running Anthropic's Claude as a quality check on OpenAI's GPT inside its own flagship product. Both companies are investors and partners, and now their models are being pitted against each other in the same pipeline. Critique and a Claude-powered Copilot Cowork agent are available now to Microsoft 365 Copilot licensees. Using one AI to fact-check another AI - inside the same product, from competing companies. Microsoft just turned model rivalry into a feature.