Details
- Microsoft AI announced seven new models at Build: reasoning, coding, image, transcription, and voice systems built as one family.
- The lineup includes MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Image-2.5-Flash, MAI-Transcribe-1.5, MAI-Voice-2, and MAI-Voice-2-Flash.
- Microsoft says the models were built from scratch on a clean data lineage and designed for efficiency rather than distilled from other systems.
- MAI-Code-1-Flash is positioned as a faster coding model that outperforms Claude Haiku 4.5 on Microsoft’s tested coding benchmarks while using up to 60% fewer tokens.
- MAI-Transcribe-1.5 is described as a speed-and-accuracy leader on Microsoft’s cited speech benchmarks, while MAI-Voice-2 adds emotion control, speaker consistency, and code-switching.
- Microsoft says the multimodal models are available in Microsoft Foundry, with some also live on OpenRouter and the reasoning model coming soon to additional partners.
Impact
Microsoft is broadening its AI stack with in-house models that target core enterprise use cases across code, speech, images, and reasoning. That reduces reliance on outside model providers inside its own products and gives Microsoft more control over cost, latency, and deployment options. The coding and transcription claims also suggest a direct push against frontier-model vendors in developer workflows, while the staged availability across Foundry and partner platforms should help drive adoption.
