Microsoft shifts to in-house MAI models to cut AI costs, impacting developers and pricing

Microsoft is shifting away from OpenAI and Anthropic models in favor of its own MAI model family to cut costs, according to a Bloomberg report. Tens of thousands of prompts in Excel and Outlook are now handled by MAI models, though Copilot still processes millions weekly. For AI builders, this signals a strategic pivot toward cost-efficient in-house inference that could reshape pricing and model availability on Azure.

What happened

Microsoft has begun replacing OpenAI and Anthropic models with its in-house MAI stack in a growing set of applications, including Excel and Outlook, according to a Bloomberg report citing a person familiar with the company's AI strategy. The move is driven by rising costs of top-tier models from OpenAI and Anthropic. Microsoft AI CEO Mustafa Suleyman 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."

At Build 2026, Microsoft unveiled seven MAI models, including its first reasoning model, MAI-Thinking 1. It is a midsized model with 35 billion active parameters and a 256,000-token context window. In blind tests, it matched the coding capabilities of Anthropic's Claude Opus 4.6. Other MAI models cover image generation, transcription, voice recognition, and coding tasks.

The cost pressure is industry-wide. OpenAI's GPT-5.5 costs $5 per million input tokens and $30 per million output tokens. Anthropic's Fable 5 costs $10 per million input tokens and $50 per million output tokens. By contrast, DeepSeek's V4-Pro charges just $0.435 per million input tokens and $0.87 per million output tokens. Companies like Amazon, Accenture, Meta, and Uber are reportedly seeking cheaper AI options.

Why AI builders should care

This shift highlights a growing emphasis on cost efficiency and in-house inference over reliance on external providers for core productivity tools. For developers building on Microsoft's ecosystem, the MAI model family offers a direct alternative to OpenAI and Anthropic models on Azure, potentially lowering inference costs and providing more control over data residency. The broader market trend toward reducing AI bills suggests that price sensitivity is becoming a key factor in model selection.

Practical implications

As MAI rolls out across more apps, developers may encounter trade-offs in cost, performance, and model choice. MAI-Thinking 1's coding parity with Claude Opus 4.6 suggests that in-house models can compete on quality while reducing token costs. However, the exact pricing for MAI models has not been disclosed, and the OpenAI deal remains through 2032, so the transition will be gradual. Developers should monitor Azure AI model availability and pricing updates to plan their inference strategies.

Caveats

The details of the MAI rollout are based on internal strategy discussions reported by Bloomberg and may not reflect final configurations. Microsoft publicly asserts that its MAI models are not as sophisticated as leading frontier AI systems, so the shift may be limited to specific use cases where cost savings outweigh capability differences. The OpenAI partnership continues through 2032, meaning Microsoft will maintain access to those models even as it accelerates internal development.