Microsoft Bets M365 on Multi-Model: GPT Drafts, Claude Reviews, and the Single-Model Era Is Over

Microsoft announced Critique on March 30. Here's how it works: when you use M365 Copilot Researcher, GPT drafts the initial research response. Then Claude reviews it for accuracy, completeness, and citation quality. You only see the final result after both models have had their pass.

Microsoft claims a 13.8% improvement on the DRACO benchmark, which translates to +7.0 points over Perplexity Deep Research running Claude Opus 4.6 alone.

There's also a "Council" mode that shows multiple model responses side-by-side with a cover letter explaining where they agree and where they diverge. Plans for bidirectional critique are coming, meaning Claude would draft and GPT would review.

This caught me off guard. Not the technology, the draft-then-critique pattern is obvious to anyone who's built agent pipelines. What surprised me is Microsoft shipping it as a first-class feature in M365. This is the clearest signal yet that enterprise AI is moving from "pick a model" to "orchestrate models." The single-vendor era lasted about 18 months.

For builders, the pattern is directly implementable today. Use one model for generation, another for verification. I've been doing this in my own workflows, using Haiku for fast drafts and Opus for review, and the quality difference is noticeable. Microsoft just validated it at enterprise scale.

The competitive implication is interesting too. Microsoft is essentially saying "GPT alone isn't enough for our flagship product." That's a remarkable admission. And it positions Anthropic as the verification layer, the trust arbiter, which might be a more valuable position than being the generator.

Composio open-sourced Agent Orchestrator the same week, billing it as "the coordination layer that turns AI coding agents from a toy into a production system." The multi-model orchestration pattern is converging fast.