Microsoft Replaces OpenAI and Anthropic Inside Office Apps

Microsoft has started routing tens of thousands of weekly AI prompts inside Excel and Outlook to its own home-grown “MAI” models, Bloomberg reported Tuesday, the clearest public sign yet that the software giant is willing to swap out its frontier-lab partners to slash its own AI bill. The shift lands two days after Microsoft cut roughly 4,800 jobs, or 2.1% of its global workforce, in a wave that hit Xbox and commercial sales hardest.

The pivot marks a quiet turn for a company that has spent the past three years running most of its Copilot features on OpenAI’s GPT models and adding Anthropic’s Claude to its model picker last year. Microsoft AI chief executive Mustafa Suleyman said the goal in the long run is to “reduce and ultimately eliminate” what Microsoft pays Anthropic. The in-house models still handle only a small fraction of Microsoft’s total AI footprint. The direction is now explicit.

Office Prompts Quietly Move to Microsoft’s Own Models

Bloomberg’s report, republished by PYMNTS on Tuesday, said Microsoft’s MAI models are now processing tens of thousands of requests per week inside Excel and Outlook. Both apps previously leaned more heavily on models from OpenAI and Anthropic. Microsoft routing Office prompts to in-house MAI models covers only a small share of total traffic so far.

The MAI models already cover a portion of GitHub Copilot for software developers. Suleyman confirmed a Microsoft-built transcription model is on track to integrate into Teams and other products in the coming months. A Microsoft spokesperson, asked by PYMNTS about the Bloomberg report, declined to comment. The company told reporters earlier this year it is “using a mix of models, which includes models from OpenAI as part of our partnership, as well as Microsoft AI and open-source models.” That mix is now tilting toward the in-house stack.

Anthropic, Not OpenAI, Is the Target

Suleyman told Bloomberg in an interview that Anthropic, not OpenAI, is the cost line Microsoft is most eager to close. “Anthropic is extremely expensive and I think many people are urgently looking for alternatives,” he said, in remarks covered by The Next Web. Suleyman’s interview about cutting the Anthropic bill lays out a strategy that mirrors the Microsoft AI roadmap.

That asymmetry has a structural reason. Microsoft holds discounted access to OpenAI models through 2032 under a renegotiated partnership that, until late 2025, had barred Microsoft from independently building frontier AI. Anthropic sits outside that arrangement. PYMNTS reported in January that Microsoft had become one of Anthropic’s top customers, on track to spend around $500 million per year running Anthropic’s models inside Microsoft products. The discount it enjoys on OpenAI does not extend to Anthropic.

Suleyman’s argument arrives as enterprise AI spending becomes a margin problem for Microsoft. After refining the MAI models for consulting firm McKinsey, he said Microsoft was able to outperform OpenAI’s GPT 5-5 with ten times better cost efficiency. The company is also in talks with Adobe about adopting the same models, Bloomberg reported. How Claude Code and Codex compare on agentic coding is the contest Microsoft is now trying to undercut on price.

Inside the Seven-Model MAI Slate

At its annual Build developer conference this week, Microsoft unveiled seven new in-house AI models under the MAI brand. The headliner is MAI-Thinking-1, a reasoning model Microsoft says matches Anthropic’s prior-generation Opus 4.6 on a widely used coding benchmark. Microsoft also showed refreshed image, voice, and transcription models, plus MAI-Code-1, a five-billion-parameter coding assistant tuned for GitHub Copilot and VS Code.

MAI-Thinking-1 is a mid-sized reasoning model with 35 billion active parameters and a 128K context window, according to a breakdown by The AI Economy’s Ken Yeung. MAI-Thinking-1’s specs and how the seven models compare detail Microsoft’s entry into the reasoning-model category. Suleyman said the model achieved 97% on AIME 2025 and 53% on SWE-bench Pro, the latter figure the company claims places it alongside Opus 4.6 on the toughest coding benchmark in wide use.

Microsoft’s claim is contested. The MAI-Thinking-1 benchmark gap against frontier labs is real, with the benchmarks released at Build showing Thinking-1 trailing the latest OpenAI and Anthropic models by a wide margin and landing roughly on par with Deepseek V3.2. The Decoder also reported that the model trails Sonnet 4.6 and Opus 4.6 in coding on the public benchmarks, a gap that Microsoft’s human-evaluation marketing does not, on the record, dispute.

Suleyman said the full slate is built around “real attention to detail and a commitment to making very practical and efficient tools that are tuned to just how you work in the real world.” Microsoft trained the MAI models on what it describes as clean, commercially licensed data, a position The Decoder noted sits in tension with the company’s use of the Common Crawl dataset. The company has framed the broader lineup as part of a “humanist superintelligence” vision that puts people ahead of raw capability.

1. August 2025: MAI-Voice debuts in Copilot Daily, Podcasts, and Copilot Labs, generating a minute of audio in under a second on a single GPU.
2. Earlier 2026: MAI-Image-1 enters the top 10 of LMArena’s text-to-image leaderboard.
3. March 2026: MAI-Image-2 ships with a photorealism focus and readable in-image text.
4. April 2026: MAI-Image-2-Efficient and MAI-Transcribe launch on Foundry; MAI-Voice reaches commercial availability on Foundry.
5. Build 2026: Seven new MAI models ship, including MAI-Thinking-1, MAI-Code-1, and refreshes across image, voice, and transcription.

Two Days Earlier, Microsoft Cut 4,800 Jobs

The Bloomberg AI report landed two days after Microsoft cut 4,800 jobs on Monday, July 6, roughly 2.1% of its global workforce. TechCrunch’s reporting on the cuts, drawn from internal memos, names Xbox and commercial sales as the two areas absorbing the most pain. Microsoft’s 4,800-person layoff and Xbox restructure is the largest single-day cut at the company since its 2025 layoffs.

Xbox absorbed 1,600 of the cuts immediately, with another roughly 3,200 expected through fiscal year 2027. Xbox CEO Asha Sharma called it “the most significant restructure in Xbox history.” Sharma’s own memo, obtained by The Guardian, said Xbox “is operating at margins that are 3-10 times lower than comparable platform and publishing businesses.” Four Xbox studios will leave as part of the restructure. Compulsion Games and Double Fine Productions will become independent. Ninja Theory and Undead Labs have entered terms to join new owners. Arkane’s management in France is entering a required works-council consultation over what Sharma called “potential strategic options.”

Amy Coleman, Microsoft’s executive vice-president and chief people officer, wrote in a company-wide memo that the eliminated roles “are not being replaced by AI.” She added that “AI is changing how work gets done” and that “some of the tasks we do every day can now be automated.” Coleman said Microsoft has redeployed more than 4,000 employees into new roles over the past year, including another 500 this month. The Guardian’s reporting, drawing on the same memo, said the commercial-side cuts will build on a $2.5 billion push to embed 6,000 engineers inside enterprise clients.

The cuts continue a pattern that has run for more than a year. Microsoft laid off roughly 15,000 employees across two rounds in 2025, per TechCrunch. The latest round comes against an industry backdrop in which close to 154,000 tech workers have lost their jobs in the first half of 2026, with Meta, Oracle, Amazon, and Cognizant cutting thousands of workers of their own.

Sharma succeeded Phil Spencer as Xbox chief in February, after Spencer retired, per The Guardian. The same memo said Sharma has pledged to return the Xbox division to growth by 2027.

• 4,800: jobs cut on July 6, 2026
• 2.1%: share of Microsoft’s global workforce
• 1,600: Xbox roles eliminated immediately
• 3,200: further cuts expected through fiscal year 2027
• 15,000: Microsoft layoffs across 2025 in two rounds
• 154,000: tech industry job losses in the first half of 2026

The $2.5 Billion Frontier Company Bet

The same week as the layoffs, Microsoft formally stood up a new business unit called Frontier Company. The unit pairs Microsoft’s enterprise AI tools with 6,000 forward-deployed engineers and is backed by a $2.5 billion investment, a figure confirmed both in Coleman’s layoff memo and in The Guardian’s reporting. Microsoft’s 6,000-engineer AI deployment push is the operational shape of the strategy the MAI models are meant to serve.

The unit is meant to embed Microsoft engineers inside enterprise customers to push AI adoption in companies that have so far been slow to move. Coleman framed it in her memo as a way to “adjust resources and roles and shift how we operate so we can have the greatest impact for our customers.” The same week, Microsoft announced four gaming studios would leave Xbox and a fifth was entering review.

Suleyman has been direct about the cost calculus behind the model rollout. In remarks at Build and in the Bloomberg interview that followed, he put a number on what Microsoft is trying to wind down.

We pay a lot of money to Anthropic, so our goal is to reduce and ultimately eliminate that cost.

Suleyman said those words at Build in June. Anthropic has already shipped two more advanced models since Opus 4.6, per The Next Web’s reporting, giving the lab a lead of several months. Whether the MAI lineup can match Anthropic’s latest generation, rather than a prior one, is the test for the seven new models.

What Office Customers Will Actually Get

For Microsoft 365 and Copilot customers, the shift may show up first as a pricing change. Satya Nadella has hinted that AI billing could shift toward usage-based pricing rather than flat-rate subscriptions, per The Decoder’s coverage of Build. One setup floated in that report would make the cheaper MAI models the default, with OpenAI and Anthropic available as premium add-ons at extra cost. The change would, in effect, push Microsoft’s third-party AI bill onto customers who want the older frontier models.

The risk inside that plan is straightforward. MAI-Thinking-1 trails the latest OpenAI and Anthropic models on the public benchmarks, landing roughly on par with Deepseek V3.2. If Microsoft quietly defaults Office customers to MAI, the same subscription price could mean weaker AI inside the apps customers already pay for. Microsoft has not committed publicly to any specific default. Suleyman’s MAI Superintelligence team, formed in November 2025, has shipped its first public models within six months. Catching the latest Anthropic release and staying there are different problems.

Frequently Asked Questions

Is Microsoft dropping OpenAI entirely?

No. Microsoft retains discounted access to OpenAI models through 2032 under a renegotiated partnership. What is changing is that Microsoft is now routing more of its own AI traffic to its home-grown MAI models and trying to phase out what it pays Anthropic.

What is MAI-Thinking-1?

MAI-Thinking-1 is Microsoft’s first in-house reasoning model, unveiled at Build 2026 with 35 billion active parameters and a 128K context window. Microsoft says it lands at 53% on SWE-bench Pro, which the company claims places it alongside Anthropic’s prior-generation Opus 4.6 coding model.

How much does Microsoft pay Anthropic?

PYMNTS reported in January that Microsoft had become one of Anthropic’s top customers and was on track to spend around $500 million per year on Anthropic models inside Microsoft products. Microsoft AI CEO Mustafa Suleyman has said the goal is to “reduce and ultimately eliminate that cost.”

Why did Microsoft lay off 4,800 people this week?

Microsoft cut 4,800 roles, or 2.1% of its global workforce, on Monday, July 6, 2026, with Xbox absorbing 1,600 cuts immediately and another roughly 3,200 expected through fiscal year 2027. Chief people officer Amy Coleman said the eliminated roles are “not being replaced by AI,” though she acknowledged that AI is “changing how work gets done.”

What is the $2.5 billion Frontier Company push?

Frontier Company is a new Microsoft business unit pairing enterprise AI tools with 6,000 forward-deployed engineers, backed by a $2.5 billion investment announced the week of the layoffs. The unit is meant to embed Microsoft engineers inside enterprise customers to drive AI adoption.