Microsoft launches seven new MAI models

Microsoft has launched a new suite of self-developed MAI models, including the MAI-Thinking-1 inference model, the MAI-Image-2.5 image model, the MAI-Image-2.5-Flash high-efficiency image model, the MAI-Voice-2 speech model, the MAI-Voice-2-Flash high-efficiency speech model, the MAI-Transcribe-1.5 speech-to-text model, and the MAI-Code-1-Flash encoding model. All models were trained from scratch, without distillation, and share a unified data specification and evaluation framework.