Microsoft Releases Three Proprietary AI Models on Azure to Reduce OpenAI Reliance; Agentic Stack Criticized for Complexity

Microsoft released three in-house AI models — MAI-Transcribe-1 (billed as the most accurate transcription model), MAI-Voice-1 (setting a new standard for natural speech), and MAI-Image-2 — exclusively on its Microsoft Foundry platform within Azure, signaling intent to reduce dependence on OpenAI. Azure AI Foundry also released model distillation pipelines allowing developers to fine-tune smaller models using GPT-4.5 Turbo outputs, targeting cost reduction for high-volume inference, with GitHub Copilot enterprise workflow integration at launch. However, a Forbes analysis found that Microsoft's layered agentic developer stack — spanning Copilot Studio, Azure AI Foundry, and Semantic Kernel — is confusing developers while competitors offer more streamlined APIs, posing a competitive risk. Anthropic also hired Microsoft's Eric Boyd as head of infrastructure.