Microsoft's research assistant can now use multiple AI models simultaneously
TL;DR
Microsoft Copilot Researcher now combines OpenAI GPT and Anthropic Claude in a single workflow – GPT generates initial responses, which Claude then refines.
Key Points
- The new 'Critique' feature is part of the Researcher tool in Microsoft 365 Copilot, built for complex, multi-step tasks.
- Microsoft describes the architecture as a feedback loop improving factual accuracy, analytical depth, and presentation quality.
- The feature launched alongside the general availability announcement of Copilot Cowork.
- Microsoft claims Researcher with Critique scores measurably higher on benchmarks compared to the standalone version.
Nauti's Take
One GPT drafts, one Claude critiques – sounds like a decent editorial team. Microsoft is signaling that 'which model is best?
' may have been the wrong question all along. The smarter question is how to chain models so they improve each other's output.
Whether the benchmark gains translate to real-world value is still an open question, but the architectural choice is strategically sound: no single-vendor dependency, and full flexibility to swap models as the landscape shifts.
Sources