# Microsoft trains sales staff to promote in-house AI over OpenAI and Anthropic > Source: > Published: 2026-07-16 00:06:29+00:00 # Microsoft trains sales staff to promote in-house AI over OpenAI and Anthropic Seven new MAI models, including a 35-billion-parameter flagship, are now being positioned as cheaper and faster alternatives to GPT and Claude. Microsoft built a multi-billion-dollar relationship with OpenAI. Now it is quietly teaching its salespeople to sell around it. At its Build 2026 conference in early June, Microsoft unveiled seven new in-house models under the MAI banner, short for Microsoft AI. The flagship, MAI-Thinking-1, runs on 35 billion parameters with a 256K context window. The pitch to enterprise customers: it does what GPT and Claude do, at a fraction of the cost. That is not a small claim. Microsoft says its MAI models can deliver up to 10 times better cost efficiency in tuned enterprise workloads compared to competing models. The company also says MAI-Thinking-1 can match or outperform GPT-5.5 and Claude Opus 4.6 on key benchmarks, particularly in coding tasks measured by SWE-Bench Pro. ## From partner to competitor, sort of Microsoft holds a stake in OpenAI valued at roughly $135 billion as of late 2025. Starting July 7, Microsoft began migrating selected Microsoft 365 applications, specifically Excel and Outlook, away from OpenAI and Anthropic models and onto its own MAI stack. The company is now routing tens of thousands of prompts per week through its own models instead of paying inference costs to outside providers. AI inference costs have become one of the largest and fastest-growing line items for any company deploying AI at enterprise scale. By swapping in cheaper internal models for high-volume, lower-complexity tasks like spreadsheet formulas and email drafting, Microsoft can meaningfully reduce its operating costs without compromising the user experience it delivers through Copilot and other products. ## What the benchmarks actually show Independent evaluations gave MAI-Thinking-1 a credible performance record. In blind tests, human raters preferred it over Anthropic’s Claude Sonnet 4.6. On coding benchmarks, it delivered results competitive with Claude Opus 4.6, which sits at the top of Anthropic’s model lineup. The 256K context window on MAI-Thinking-1 means the model can process and respond to much longer inputs in a single query, allowing it to handle a lengthy legal contract or a full earnings report without chopping the document into pieces. ## What this means for the broader AI market For OpenAI and Anthropic, the implications are uncomfortable. Microsoft has historically been one of the most significant distribution channels for both companies. Azure hosts OpenAI models. Microsoft 365 Copilot was built, in large part, on OpenAI’s infrastructure. If Microsoft routes a meaningful portion of that volume to its own models, the revenue and usage signal flowing to OpenAI changes materially. Anthropic’s Claude models, particularly the Sonnet and Opus tiers, are central to its enterprise revenue story. If MAI-Thinking-1 is beating Claude Sonnet 4.6 in independent evaluations and being sold actively to Microsoft’s enterprise customer base, that is a direct challenge to Anthropic’s positioning in exactly the market segment it has been targeting most aggressively. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).