Microsoft releases Mage, a 4B AI model

Microsoft released Mage, a multimodal model family built to a fixed 4B-parameter budget: Mage-VL for codec-native image and video understanding, and Mage-Flow, pairing a VAE with a native-resolution diffusion transformer for text-to-image generation and instruction-based editing. Microsoft says both train, fine-tune, and deploy on modest hardware while remaining competitive with much larger open systems.