Microsoft Revamps Copilot With Anthropic's Claude — New Multi-Model 'Critique' Feature Launches in Wave 3

Microsoft unveiled a significant overhaul of Microsoft 365 Copilot on March 31, 2026, anchored by a new "Critique" feature that runs two frontier AI models in sequence within the same workflow. Under the system, OpenAI's GPT drafts an initial response — whether a document summary, research memo, or code block — and Anthropic's Claude then reviews the output for completeness, factual accuracy, and citation integrity before the result reaches the user. The company framed the dual-model approach as a measurable step toward reducing AI hallucinations for enterprise use cases where accuracy is critical, citing internal testing showing a 37% reduction in factual errors on document-heavy tasks compared to single-model outputs.

The Critique feature is the centerpiece of "Wave 3" of Microsoft 365 Copilot, which also brings Anthropic's Claude models into mainline Copilot Chat for the first time — expanding beyond the specific integrations where Claude was previously available. Microsoft described the underlying architecture as a "multi-model mesh" in which different AI systems are assigned different roles based on their comparative strengths: GPT's speed and instruction-following excel at first-pass drafting, while Claude's strength in long-document analysis and citation checking makes it the preferred reviewer. New enterprise pricing tiers introduced alongside Wave 3 reflect the added compute cost of running sequential model calls.

The deeper integration follows Microsoft's $30 billion Azure compute deal with Anthropic struck in November 2025, which gave Anthropic substantial cloud capacity in exchange for Microsoft becoming a preferred distribution partner for Claude. Analysts at Goldman Sachs noted that the Wave 3 launch positions Microsoft as a neutral "AI orchestration platform" rather than an OpenAI-exclusive partner — a strategic hedge against the risk of any single frontier model losing its performance lead. The move also coincides with a broader industry shift: enterprise AI infrastructure is increasingly converging around multi-model workflows, with Salesforce, SAP, and ServiceNow all announcing similar multi-vendor AI routing architectures in Q1 2026.