Quick Facts
- Microsoft’s MAI models now complete tens of thousands of AI prompts weekly in Excel and Outlook, displacing OpenAI and Anthropic models.
- Microsoft AI CEO Mustafa Suleyman said the company aims to “reduce and ultimately eliminate” its Anthropic spending, calling the service “extremely expensive.”
- Microsoft projects a 10x improvement in output tokens per dollar from its fine-tuned MAI models compared to GPT-5.5.
Microsoft is pulling back from OpenAI and Anthropic models inside its core productivity products and replacing them with its own Microsoft AI (MAI) model family, according to a Bloomberg report published July 7, 2026.
Tens of thousands of AI prompts in Excel and Outlook are now completed each week using MAI models. Those products previously relied more heavily on OpenAI and Anthropic.
The Cost Math
The financial case is direct. OpenAI charges $5 per million input tokens and $30 per million output tokens for GPT-5.5. Microsoft accesses those models at a discount through its partnership, but the volume is enormous. Anthropic’s pricing runs higher still.
Microsoft AI CEO Mustafa Suleiman told Bloomberg: “Anthropic is extremely expensive and I think many people are urgently looking for alternatives. We pay a lot of money to Anthropic, so our goal is to reduce and ultimately eliminate that cost.”
Microsoft began winding down most internal Claude Code licenses in mid-May 2026. Microsoft shares rose 2% after the news broke.
The MAI Model Lineup
At its Build 2026 developer conference on June 2, Microsoft introduced seven MAI models spanning reasoning, coding, image generation, speech, and transcription.
The flagship MAI-Thinking-1 uses a 35-billion active-parameter mixture-of-experts architecture with a 256,000-token context window. It scores 97.0% on AIME 2025 and 94.5% on AIME 2026. Microsoft says it matches Claude Opus 4.6 on SWE-Bench Pro at lower cost.
Both MAI-Thinking-1 and MAI-Code-1-Flash were trained end-to-end on commercially licensed data with no distillation from OpenAI, Anthropic, or any third-party model. MAI-Code-1-Flash is already the default in the Copilot auto router inside Visual Studio Code for all paying plans.
Microsoft’s proprietary transcription model is expected to ship in Teams soon. Suleiman said in-house speech models will roll out progressively across Teams video conferencing and other products.
Partnership Restructuring Cleared the Path
An amended Microsoft-OpenAI agreement announced April 27, 2026 removed the restriction on Microsoft building its own frontier models. The deal eliminated Microsoft’s revenue share payments to OpenAI and converted its IP license from exclusive to non-exclusive through 2032.
The previous arrangement gave Microsoft exclusive access to OpenAI models until OpenAI achieved artificial general intelligence, a milestone with no defined timeline. The new terms freed Microsoft to compete directly with its own weights.
Microsoft’s investment in OpenAI Group PBC was valued at roughly $135 billion as of October 2025, representing about 27% on a diluted basis.
Pricing Model May Follow
CEO Satya Nadella has hinted that AI billing inside Microsoft products could shift toward usage-based pricing. One possible structure would make MAI models the default, with OpenAI or Anthropic models available as premium add-ons at extra cost.
The shift comes as Microsoft cuts 4,800 jobs, roughly 2.1% of its workforce, even while increasing capital spending on AI infrastructure. The MAI push represents an effort to capture more margin from AI features already embedded across its product suite.
Benchmarks published at Build 2026 showed MAI-Thinking-1 trailing OpenAI and Anthropic top models in several categories, despite Microsoft’s human evaluation claims of parity. The company’s own public acknowledgment is that MAI models are not as sophisticated as the leading frontier systems from its partners.
Read more: Microsoft is reportedly ditching OpenAI’s and Anthropic’s AI models in favor of its own to cut costs
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.