Microsoft Swaps OpenAI and Anthropic for In-House Models in Copilot

Microsoft is aggressively shifting its Copilot ecosystem away from third-party giants like OpenAI and Anthropic in favor of its own proprietary MAI models. This strategic pivot aims to slash massive licensing expenses across flagship products like Excel, Outlook, and GitHub Copilot.

The Shift to In-House MAI Models

Microsoft is currently executing a long-term strategy to reduce its dependency on external AI providers. According to recent reports, Microsoft's in-house MAI models are already processing tens of thousands of requests per week within Excel and Outlook—applications that previously relied heavily on OpenAI and Anthropic models.

Beyond productivity software, these proprietary models are being integrated into GitHub Copilot, and a specialized transcription model is slated for a future rollout in Microsoft Teams. Mustafa Suleyman, Microsoft’s head of AI, has been transparent about the economic driver behind this move, stating that the company aims to "reduce and ultimately eliminate" the high costs paid to Anthropic.

Performance Benchmarks vs. Marketing Claims

At the recent Build conference, Microsoft introduced seven new models, headlined by MAI-Thinking 1, its first dedicated reasoning model. While Microsoft claimed the model could match Anthropic’s Sonnet 4.6 and Opus 4.6 in coding tasks based on human evaluations, independent benchmarks paint a more cautious picture.

Technical evaluations suggest that Thinking-1 actually trails behind the leading models from OpenAI and Anthropic, landing closer to the performance levels of DeepSeek V3.2. This discrepancy raises questions for developers and enterprise users regarding whether the transition to in-house models will result in a noticeable dip in reasoning and coding capabilities.

Potential Impact on Pricing and User Experience

The move toward in-house models could fundamentally change how customers pay for AI. As Microsoft seeks to improve its margins, there is a risk that users may receive less capable "default" models for the same subscription price.

CEO Satya Nadella has hinted at a shift toward usage-based pricing. This could lead to a tiered ecosystem where Microsoft's MAI models serve as the standard offering, while high-performance models from OpenAI or Anthropic are relegated to premium, surcharge-based add-ons. Essentially, the costs previously borne by Microsoft may soon be passed directly to the consumer.

Microsoft has positioned its MAI models as "safe for business" because they are trained on clean, commercially licensed data. However, technical documentation reveals the use of the Common Crawl dataset—a massive collection of web data whose legal status for AI training remains a subject of intense debate. While this practice is common across the industry, Microsoft’s emphasis on "clean" data highlights the tension between corporate marketing and the legal realities of web-scale scraping.

Key Takeaways

  • Cost Optimization: Microsoft is replacing expensive OpenAI and Anthropic models with in-house MAI models in Excel, Outlook, and GitHub Copilot to boost margins.
  • Performance Gap: Despite claims of parity with top-tier models, early benchmarks suggest Microsoft's reasoning models currently trail behind industry leaders.
  • Pricing Evolution: Users may see a shift toward usage-based pricing, where premium third-party models become paid add-ons rather than standard features.