On July 9 (local time), OpenAI announced its latest flagship model, GPT-5.6, and said it will become the preferred model in Microsoft 365 Copilot[1]. The model is being built into everyday work tools including Word, Excel, PowerPoint, Copilot Chat, and Cowork, promising higher-quality document creation and data analysis.
GPT-5.6 Named Copilot's "Preferred Model"
According to the announcement, GPT-5.6 will serve as the new preferred model across Microsoft 365 Copilot's apps: Word, Excel, PowerPoint, Copilot Chat, and Cowork[1]. Microsoft 365 users will be able to get help from OpenAI's newest model inside the tools they already use every day.
The delivery approach is also notable. In addition to serving GPT-5.6 natively within its own services, Microsoft will bring the model to Microsoft 365 customers by accessing the OpenAI API directly[1]. Nitin Agrawal, President, Copilot & Agents Core at Microsoft, said customers will be able to produce more polished outputs, from drafting documents to analyzing data and building presentations[1].
What Changes in Each App
OpenAI outlines the following improvements across the Microsoft 365 apps[1].
In Word, users can draft, edit, and refine documents with fewer rounds of prompting. In Excel, the model supports deeper analysis while using tokens efficiently, shortening the path from data to insight. In PowerPoint, early ideas can be turned into visually polished presentations with less manual guidance. In Cowork, the collaborative tool, complex cross-functional work can be pulled together into high-quality outputs with less manual coordination.
All of these point in the same direction—"better results with less effort"—reflecting GPT-5.6's design goal of getting more useful work out of every token[1].
What Kind of Model Is GPT-5.6?
GPT-5.6 is OpenAI's newest model family, rolled out on July 9. It has a three-tier lineup: the top-end "Sol," plus the lighter, more cost-efficient "Terra" and "Luna," covering everything from flagship tasks to high-volume use within a single series[2].
According to OpenAI's own figures, on the "Agents' Last Exam" benchmark—which evaluates long-horizon agent work across 55 fields—GPT-5.6 Sol set a new high of 53.6, beating Anthropic's Claude Fable 5 (adaptive reasoning) by 13.1 points. At medium reasoning, it still topped Fable 5 by 11.4 points at roughly one-quarter of the estimated cost, and the lower Terra and Luna tiers also outperformed Fable 5 at around one-sixteenth of the estimated cost[2]. Because these are all numbers published by OpenAI itself, independent third-party verification is still to come.
A "Honeymoon" Message at a Delicate Time
The announcement also comes amid continuing reports about the relationship between the two companies. Several international outlets say Microsoft is building its own AI models (MAI) for Microsoft 365 as part of managing costs. The understated "preferred model" phrasing is seen as leaving room for both realities at once: OpenAI's model can keep that spot while Microsoft advances its own models in parallel[3].
Against that backdrop, Nikunj Handa, Head of API Product at OpenAI, emphasized the continuation of the partnership, saying "Microsoft 365 is where millions of people write, analyze, create, and collaborate every day"[1]. By putting the preferred-model designation front and center, the announcement can also be read as a signal that the partnership remains healthy.
Summary
OpenAI's latest model, GPT-5.6, is being built into Microsoft 365 Copilot's Word, Excel, PowerPoint, Copilot Chat, and Cowork as the preferred model. With its token-efficient design, it aims to deliver "high-quality results with less effort" inside everyday work tools. Coming at a delicate moment marked by reports of Microsoft building its own models, it is a story worth watching—including for where the two companies' relationship heads next.
出典:https://openai.com/index/gpt-5-6-preferred-model-microsoft-365-copilot
