On Monday, Microsoft announced a new feature called “Critique” as part of its Microsoft 365 Copilot Researcher toolkit. In practice, OpenAI’s GPT model first conducts research and generates an initial draft, after which Claude reviews the output following an academic-style evaluation process — assessing accuracy, completeness, and citation quality — before producing a final report for the user.

Microsoft noted that this system is expected to become bidirectional in the future, where Claude may generate the initial draft and GPT would then review and refine it. According to the DRACO benchmark, which measures deep research quality, collaboration between models significantly outperforms single-model outputs. This approach helps reduce AI hallucinations while improving productivity and output quality.

Microsoft has also introduced a “Council” mechanism, where multiple models conduct research simultaneously and their outputs are compared. In demonstrations, GPT and Claude independently generated research reports, after which a third “judge” model evaluated the results, identified areas of agreement and divergence, and summarized each model’s unique contributions.

Both advanced features — Critique and Council — have been integrated into the Microsoft 365 Copilot Researcher toolkit and are currently in early testing. Initial access is limited to enterprise customers participating in Microsoft’s Frontier program.

The rollout of this deep research system strengthens Microsoft’s competitive moat in the enterprise productivity market and signals a broader shift in the AI industry — from a focus on model parameters to more sophisticated system integration and logical validation.

Microsoft’s latest move also highlights the evolving alliances within the AI ecosystem. In October last year, OpenAI announced a restructuring agreement with Microsoft, allowing Microsoft to retain intellectual property rights for OpenAI’s models and products until 2032, while both parties are permitted to collaborate with each other’s competitors.

Market Interpretation

In November, Microsoft announced new strategic partnerships with Nvidia and Anthropic, alongside plans to invest up to $5 billion in OpenAI competitors. As part of these developments, Microsoft introduced the Copilot Cowork agent earlier this year, incorporating Anthropic’s technology — a move that previously triggered a sell-off in U.S. software stocks.

Overall, Microsoft’s latest innovations underscore a critical industry shift: AI competition is no longer just about building bigger models, but about creating smarter, more reliable systems through collaboration, validation, and integration.