Two Microsoft Copilot stories landed within days of each other in July 2026. Microsoft said its own MAI models are handling some Copilot requests, and OpenAI said GPT-5.6 is now the preferred model in Microsoft 365 Copilot.
Key Takeaways
- OpenAI announced that GPT-5.6 is now the preferred model in Microsoft 365 Copilot. Press coverage dates Microsoft’s deployment to July 9, 2026.
- The Decoder reported on July 7 that Microsoft’s in-house MAI models are “already processing tens of thousands of requests per week in Excel and Outlook.”
- VentureBeat’s report on Microsoft’s newer MAI models lists image, voice, code and transcription models and a claimed GPU cost cut of up to 89% in Dynamics 365 Contact Center.
- The 2mind view: these are different layers, not a contradiction. MAI handles specific cost-sensitive workloads while GPT-5.6 is the preferred model for the main Copilot experience.
What happened
The Decoder reported on July 7, 2026 that Microsoft is replacing some OpenAI and Anthropic model use with its own models. It says MAI models are “already processing tens of thousands of requests per week in Excel and Outlook,” and quotes the line “our goal is to reduce and ultimately eliminate that cost” in the context of payments to Anthropic.
OpenAI then announced that GPT-5.6 is now the preferred model in Microsoft 365 Copilot. Press coverage dates Microsoft’s deployment to July 9, 2026. The coverage we could read does not define what “preferred” means in practice.
For the contract side of this relationship, see our reading of Microsoft’s April 2026 amendment with OpenAI.
Related reading: Microsoft weighing a swap of ChatGPT for Kimi K3 and who carries the Nvidia–OpenAI backstop risk.
| Signal | What the sources say |
|---|---|
| MAI in Copilot (The Decoder, Jul 7) | “Tens of thousands of requests per week” in Excel and Outlook |
| MAI model line (VentureBeat) | Image, voice, code and transcription models; claimed GPU cost cut of up to 89% in Dynamics 365 Contact Center |
| GPT-5.6 in Copilot (OpenAI, Jul 9) | Preferred model in Microsoft 365 Copilot |
Read together, the sources describe two layers. MAI models handle specific workloads where cost matters, and the VentureBeat list is mostly non-text models. GPT-5.6 is named the preferred model for the main Microsoft 365 Copilot experience.
Where the Two Lists Overlap
VentureBeat’s list of surfaces that use Microsoft’s in-house models includes Excel, PowerPoint, Copilot Chat and Outlook. Coverage of the OpenAI announcement names Word, Excel, PowerPoint and Copilot Chat as Microsoft 365 Copilot surfaces. Some outlets add other surfaces, and they do not agree on which.
| Surface | In VentureBeat’s MAI list | In coverage of the OpenAI announcement |
|---|---|---|
| Excel | Yes | Yes |
| PowerPoint | Yes | Yes |
| Copilot Chat | Yes | Yes |
| Outlook | Yes | Not named |
| Word | Not named | Yes |
Excel, PowerPoint and Copilot Chat appear on both lists. Neither source says which model handles which request inside those apps, so the overlap is a reason to ask Microsoft for routing details, not evidence of replacement.
GPT-5.6 is a family of three models, Sol, Terra and Luna, according to OpenAI’s system card. The coverage we read does not say which one Copilot uses as the preferred model, a detail that matters for cost. See our token price comparison.
The two lenses
Lens one: the “hedging in public, committing in private” lens. One interpretation is that Microsoft’s public messaging about reducing OpenAI dependency is real at the infrastructure level — home-grown MAI models are genuinely being developed and deployed in some product lines — but that at the actual application layer, where user experience and immediate product quality matter most, OpenAI’s frontier models remain simply the best available option.
Under this reading, Microsoft isn’t being inconsistent so much as pragmatic: build alternatives quietly, but don’t sacrifice product quality by forcing an inferior in-house model onto customers before it’s ready. GPT-5.6’s adoption as Preferred Model could reflect genuine technical merit rather than contradicted strategy.
Lens two: the “strategic incoherence” lens. A more skeptical reading is that this reveals real tension inside Microsoft between different teams and priorities — the infrastructure and cost-control side pushing for MAI adoption, and the product side prioritizing whatever model performs best for enterprise customers, regardless of vendor.
If that’s the case, customers and partners relying on Microsoft’s AI roadmap should expect continued unpredictability: model defaults could shift again as internal priorities shift, making long-term planning around any single Microsoft AI stack harder than it appears from outside. This lens treats the OpenAI/MAI relationship as an unresolved internal negotiation rather than a settled hierarchy.
Why it matters
This matters most to enterprise IT teams and developers building on Microsoft’s Copilot ecosystem, since the “Preferred Model” designation affects default behavior across a huge installed base of Microsoft 365 users.
It also matters to OpenAI, whose commercial relationship with Microsoft remains its most important distribution channel even as both companies pursue increasingly independent paths — OpenAI with its own infrastructure ambitions, Microsoft with its own model development.
And it matters to competitors like Anthropic and Google, who are watching how much room exists in Microsoft’s stack for alternative models when the “preferred” slot keeps going back to OpenAI.
What to watch next: whether Microsoft’s MAI models eventually earn similar “preferred” status in any Microsoft product, and whether that shift happens gradually or abruptly. That will tell us whether today’s contradiction is a transition phase or a more permanent pattern.
FAQ
Q. What is Microsoft 365 Copilot’s “Preferred Model” designation?
A. OpenAI announced that GPT-5.6 is now the preferred model in Microsoft 365 Copilot, with deployment starting July 9, 2026. The coverage we could read does not say whether that makes it the default for every request.
Q: Does this mean Microsoft is abandoning its own MAI models?
A: Not necessarily — reporting indicates MAI models are still being developed and deployed in some contexts, but GPT-5.6’s Preferred Model status shows OpenAI still holds a significant position within Microsoft’s product ecosystem.
What would change our view
If Microsoft’s MAI models earn a ‘Preferred’ designation in even one high-visibility Microsoft 365 product within two quarters, the hedging reading in Lens one needs revisiting.
That would suggest the in-house models are closing the gap faster than the current cost-driven framing implies.
We’d also revisit this if OpenAI loses the Preferred Model designation in a future Copilot update without a clear technical justification given.
Sources
- GPT-5.6 is now the preferred model in Microsoft 365 Copilot — OpenAI, July 2026
- The Decoder — 2026-07-07. Microsoft routes Excel and Outlook Copilot prompts to in-house MAI models to cut inference cost
- The Next Web — 2026-07-07. Microsoft swaps in its own AI over OpenAI in some apps
- VentureBeat — 2026-07-23. Microsoft says in-house models cut costs versus OpenAI
Leave a Reply