Microsoft enhances Copilot research tool with multi-model AI capability

Microsoft has introduced a new upgrade to its Copilot platform, aiming to improve how users conduct complex research tasks by integrating multiple artificial intelligence models. The update allows Copilot to draw on the strengths of both OpenAI and Anthropic, marking a shift towards collaborative AI systems within a single workflow.

Announced in a company blog post, the feature forms part of the Microsoft 365 Copilot ecosystem and is designed specifically for its Researcher tool. Unlike the standard Copilot assistant, Researcher focuses on handling multi-step queries that require deeper analysis and information synthesis. The new capability is intended to improve the quality and reliability of results for professional and academic use.

New critique feature improves research accuracy

At the centre of the update is a new function, Critique, that introduces a layered approach to generating responses. The system first produces an answer using models developed by OpenAI, then reviews and refines it with models from Anthropic. This process is designed to create a feedback loop that enhances the final output.

Microsoft explained the reasoning behind this approach, stating that “this architecture creates a powerful feedback loop that delivers higher-quality results across factual accuracy, analytical breadth, and presentation”. The company added that the workflow mirrors practices commonly seen in academic and professional research environments, where peer review and iterative refinement are standard.

The company also reported that the upgraded Researcher tool performs better on internal benchmarks, particularly those measuring accuracy, completeness, and objectivity. According to Microsoft, the system delivers improved results compared with competing deep learning models, including those developed by Perplexity. While these claims have not been independently verified, they highlight the growing competition among AI providers to deliver more reliable research tools.

Anthropic itself has previously explored similar concepts through its own research features, which can deploy multiple instances of its Claude model to handle complex queries. Microsoft’s implementation, however, combines models from different organisations, reflecting a broader trend towards interoperability in artificial intelligence systems.

Model council offers comparative AI insights

In addition to the Critique feature, Microsoft has introduced another option called Model Council. This tool is designed for users who prefer greater transparency and control over the research process. Rather than merging outputs into a single response, Model Council presents answers from both OpenAI and Anthropic side-by-side.

The system also generates a report highlighting areas where the models agree or differ. This allows users to evaluate contrasting perspectives and make more informed decisions based on the available information. The feature may be particularly useful for tasks that require critical analysis or for which differing viewpoints are valuable.

By offering both the Critique and Model Council options, Microsoft is positioning Copilot as a more flexible research assistant that can adapt to different user needs. Those seeking streamlined, refined answers can rely on the Critique workflow, while users who prioritise transparency can opt for comparative outputs.

Both features are currently available through the Frontier programme within Microsoft 365 Copilot. This programme serves as an early-access platform where users can try new AI capabilities before they are rolled out more widely. It also provides Microsoft with feedback that can help refine its tools further.

The introduction of multi-model collaboration reflects a broader evolution in artificial intelligence, where combining systems may deliver better outcomes than relying on a single model. As organisations continue to explore these approaches, tools like Copilot’s Researcher may play an increasingly important role in professional and academic settings.