OpenAI and Microsoft Deepen Copilot Ties With GPT-5.6
by Cheshta Upmanyu - 4 hours ago - 7 min read
Microsoft and OpenAI may be loosening the exclusivity that once defined their partnership, but the latest Microsoft 365 Copilot update shows that the relationship remains commercially important to both companies.
OpenAI announced on July 9, 2026, that GPT-5.6 is becoming the preferred model in Microsoft 365 Copilot, covering workplace products including Word, Excel, PowerPoint, Copilot Chat and Copilot Cowork. Microsoft separately confirmed that the model is available to customers starting today.
The timing is notable. Microsoft has recently been expanding its own AI model portfolio, reducing reliance on outside providers in some products and giving enterprise customers access to models from competing laboratories. Those moves have fuelled speculation that Microsoft and OpenAI are gradually moving toward a more distant relationship.
Yet the GPT-5.6 rollout suggests that, despite this strategic diversification, Microsoft still sees OpenAI’s flagship technology as central to its premium workplace AI offering.
GPT-5.6 Moves Directly Into Microsoft’s Productivity Suite
Microsoft and OpenAI said they worked together to optimise GPT-5.6 for knowledge-intensive tasks inside Microsoft 365.
Rather than presenting the model as an optional experimental addition, both companies are calling it the preferred model for Microsoft 365 Copilot. That designation indicates Microsoft expects GPT-5.6 to handle a significant share of complex Copilot requests, particularly tasks involving multiple applications, large amounts of business information or several stages of reasoning.
GPT-5.6 will be used across:
- Microsoft Word for drafting, revising and analysing documents
- Excel for working with data, formulas and business analysis
- PowerPoint for creating and refining presentations
- Copilot Chat for workplace research and assistance
- Copilot Cowork for longer, agent-like assignments
Microsoft describes GPT-5.6 as a frontier reasoning model suited to “agentic, end-to-end work,” signalling a shift from simple question answering toward AI systems that can complete multi-stage projects.
A Model Designed to Complete Longer Assignments
OpenAI says GPT-5.6 can create and run lightweight programs that coordinate tools, process intermediate results and determine which action should come next as a task develops.
That matters inside Microsoft 365 because workplace assignments rarely remain confined to a single prompt or application. A user might ask Copilot to inspect an Excel workbook, identify a trend, prepare a written explanation and then create a presentation from the findings.
GPT-5.6 is intended to manage more of that chain without requiring the user to manually guide every step. OpenAI says the model can complete tool-heavy tasks with fewer model calls, fewer tokens and less repeated instruction.
The model family includes three main variants:
| Model | Intended role | Likely Copilot use |
|---|---|---|
| GPT-5.6 Sol | Flagship model for the most complex tasks | Advanced reasoning, research and multi-stage assignments |
| GPT-5.6 Terra | Balance of performance and operating cost | Everyday business and productivity work |
| GPT-5.6 Luna | Faster and more economical model | High-volume or relatively lightweight requests |
OpenAI has described Sol as its strongest model so far, with improvements in areas including agentic coding, biology and cybersecurity. Terra and Luna are positioned as more cost-conscious alternatives for workloads that do not require the full capability of Sol.
The Announcement Pushes Back Against a Simple “Breakup” Narrative
The Microsoft–OpenAI relationship has changed considerably from the highly exclusive alliance that helped launch the current generative AI boom.
In April 2026, the companies amended their agreement. Microsoft’s licence to OpenAI’s intellectual property now continues through 2032 but is no longer exclusive. OpenAI can offer its products through other cloud providers, while Microsoft remains its primary cloud partner under revised terms. Microsoft also stopped paying OpenAI a revenue share.
Those changes gave both companies more freedom.
OpenAI gained the ability to distribute its products more broadly instead of depending almost entirely on Azure. Microsoft gained room to develop its own models, work with competing AI laboratories and reduce the risk of building Copilot around a single supplier.
The revised arrangement therefore resembles a strategic partnership between increasingly independent companies rather than the tightly integrated alliance seen in the early ChatGPT era.
However, describing it as a complete breakup would be misleading. Microsoft still has long-term access to OpenAI technology, OpenAI models remain deeply embedded in Microsoft products, and GPT-5.6 is now being given preferred status in one of Microsoft’s most important enterprise platforms.
Microsoft Is Still Building Alternatives Behind the Scenes
While Microsoft is adopting GPT-5.6, it is simultaneously strengthening its own AI capabilities.
The company has spent years developing internal models and testing alternatives from companies including Anthropic, Meta, DeepSeek and xAI. Reuters reported in 2025 that Microsoft was working on reasoning models that could compete with OpenAI and potentially be offered to outside developers.
More recently, reports indicated that Microsoft had started replacing external models with its own technology in selected Copilot features, including certain functions in Excel and Outlook. The apparent objective is to lower inference costs and gain greater control over the technology supporting its products.
This does not necessarily conflict with the decision to use GPT-5.6.
Microsoft can reserve a powerful OpenAI model for demanding tasks while routing simpler requests to cheaper internal models. Such a model-routing approach would allow the company to balance output quality, speed and operating cost instead of using the same expensive system for every Copilot interaction.
Copilot Is Becoming a Multi-Model Platform
Microsoft’s broader strategy increasingly appears to be based on model choice rather than loyalty to one AI laboratory.
Anthropic models have already been introduced into parts of Microsoft 365 Copilot, while Microsoft Foundry offers businesses access to a wider collection of proprietary and open models. The company has also encouraged enterprises to select or customise models according to their data, security requirements and business workflows.
GPT-5.6 can therefore be both the preferred model and one component of a larger system.
“Preferred” does not necessarily mean that every Microsoft 365 Copilot request will always be processed by GPT-5.6. Microsoft can dynamically select a model depending on the complexity of the task, the required response speed, customer settings, availability and cost.
This approach could become especially important as Copilot evolves from a chat interface into a collection of agents capable of performing work across documents, email, spreadsheets, meetings and corporate systems.
GPT-5.6 Gives Both Companies Something They Need
For Microsoft, the rollout provides an immediate capability upgrade for Copilot at a time when the company is under pressure to demonstrate that its large AI investments can produce useful enterprise products.
For OpenAI, preferred-model status places GPT-5.6 inside software used daily by businesses around the world. It also reinforces OpenAI’s position in enterprise AI as competition from Anthropic, Google, xAI and Microsoft’s own models intensifies.
The arrangement is therefore mutually beneficial even as the companies become more independent.
Microsoft gets access to OpenAI’s newest flagship technology without giving up its multi-model strategy. OpenAI receives valuable distribution through Microsoft 365 without remaining exclusively tied to Microsoft’s cloud and software ecosystem.
The Partnership Is Evolving, Not Disappearing
The GPT-5.6 announcement does not erase the tensions that have surrounded Microsoft and OpenAI.
Their agreement is less exclusive, Microsoft is building competing models, and OpenAI is expanding across other cloud platforms. Both companies are also competing more directly in workplace agents and enterprise software.
Still, placing GPT-5.6 at the centre of Microsoft 365 Copilot shows that technological and commercial cooperation remains strong where their interests overlap.
The more accurate interpretation is not that Microsoft and OpenAI are suddenly back together or on the verge of a complete separation. Their relationship is becoming more transactional, flexible and competitive.
Microsoft no longer wants Copilot to depend entirely on OpenAI. But when it needs a leading model for demanding workplace tasks, it is still prepared to put OpenAI’s technology first