Microsoft is introducing a dedicated small language model for Outlook search to improve the accuracy and relevance of email retrieval by optimizing search result ranking.
It is reported that the model uses a streamlined reranker architecture and is trained under the supervision of a larger model, optimizing search results while controlling computational resource consumption. Microsoft is currently testing the model internally and says early offline test results are encouraging.
Unlike AI search that can understand natural language instructions, this model is mainly responsible for reranking existing search results and does not involve natural language search capabilities. A test comparison shared by Microsoft shows that the new model can improve search result relevance and reduce interference from irrelevant emails.
In recent years, Microsoft has continued to integrate AI into Outlook, including features such as Copilot email drafting and automatic meeting conflict handling. Compared with generative AI that directly faces users, this search model focuses more on underlying retrieval efficiency, using lightweight technology to optimize high-frequency use cases. It also reflects that Microsoft is extending AI capabilities from content generation to performance optimization of traditional software features.
