Microsoft adds alternative AI models

Microsoft used its Build conference on June 2 to widen the set of AI models it offers developers, adding new in-house systems and more third-party options through Microsoft Foundry. The move gives customers alternatives to OpenAI models inside Microsoft’s cloud stack and comes as companies press vendors on inference costs, model choice and deployment controls. Microsoft said the additions span reasoning, coding, image, voice and transcription tasks, alongside new tooling for hosting, evaluation and governance. CNBC reported the announcements as part of a broader effort to reduce reliance on OpenAI and lower costs for developers. ### Which models did Microsoft actually add? Microsoft AI said in a keynote transcript published June 2 that it introduced “a family of seven new models” across image, voice, transcription, thinking and coding. The lineup includes MAI-Image-2.5 and MAI-Image-2.5 Flash, MAI-Transcribe-1.5, MAI-Voice-2 and Voice-2-Flash, plus new thinking and coding models referenced in Build coverage. (devblogs.microsoft.com) Nick Brady, a senior program manager for developer experience, wrote in Microsoft’s Foundry blog that “model and compute options expanded” at Build with new MAI models, Fireworks AI on Foundry, managed compute, fine-tuning and Frontier Tuning. Microsoft’s Build page said the conference was centered on AI-powered tools and platforms for developers. (microsoft.ai) ### Why does this matter if Microsoft still works with OpenAI? Microsoft in March described Microsoft 365 Copilot as “model diverse by design” and said Copilot uses leading models from OpenAI and Anthropic in an “open, heterogenous environment.” That language showed Microsoft was already positioning its AI products around multiple model suppliers rather than a single provider. (devblogs.microsoft.com) CNBC reported on June 2 that the new Build announcements were intended to lessen Microsoft’s reliance on OpenAI and lower costs for developers. Microsoft’s own product language has emphasized customer choice, performance and flexibility, while keeping OpenAI models available in its platform. ### Where do these models show up for customers? (blogs.microsoft.com) Microsoft said the new models are landing in Foundry, the company’s platform for building and running AI systems, and some are also being integrated into Microsoft products. In the keynote transcript, Microsoft said MAI-Image-2.5 models were live in PowerPoint, rolling out to OneDrive, and “landing on Foundry,” while MAI-Transcribe-1.5 was being integrated into Copilot, Teams, GitHub and Dynamics 365 Contact Center. (cnbc.com) The Foundry blog said Build 2026 added more of the components developers need for production agents, including runtime, tools, memory, grounding, models, observability and governance. Microsoft also said Foundry IQ broadened its knowledge features and that Toolboxes in Foundry entered public preview. ### What did Microsoft say about cost? Microsoft AI said in its keynote transcript that the new MAI models were built as “practical, efficient tools” and described MAI-Image-2.5 models as arriving on Foundry with “market-leading quality per dollar.” It said MAI-Transcribe-1.5 was optimized for real-world use and could produce transcripts “up to 5x faster than rival models.” (devblogs.microsoft.com) (microsoft.ai) Those claims align with Microsoft’s broader Build message around giving developers more inference and customization options. Brady’s Foundry post said the new releases were aimed at helping teams experiment with models and custom inference through Azure endpoints with enterprise controls. ### What comes next in the rollout? Microsoft said on June 2 that hosted agents in Foundry Agent Service are expected to reach general availability by early July 2026. (microsoft.ai) The company also said Foundry agents publishing to Microsoft Teams and Microsoft 365 Copilot were planned for general availability in June 2026, putting the next milestone within weeks of the Build announcements. (devblogs.microsoft.com)