What happened
Microsoft is reportedly replacing OpenAI and Anthropic AI models with its own MAI family in products like Excel and Outlook, routing tens of thousands of prompts weekly through MAI models to reduce costs. This follows Microsoft's June unveiling of seven MAI models, including MAI-Thinking 1, a 35-billion-active-parameter model with a 256,000-token context window, which matched Anthropic's Claude Opus 4.6 in blind coding tests. Microsoft AI CEO Mustafa Suleyman stated the company's goal is to reduce and ultimately eliminate Anthropic costs.
Why it matters
Cost pressures on AI model consumption are driving platform engineering decisions. Procurement teams face rising expenses for frontier models; Anthropic's Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens, significantly higher than DeepSeek V4 Pro. This follows a broader industry trend of soaring AI costs, forcing re-evaluation of build-versus-buy strategies for core AI capabilities. Platform engineers must assess internal model performance against third-party costs, prioritising efficiency over raw capability for specific workloads. Assume internal model development will accelerate; run cost-benefit analyses on all external API calls.
