Microsoft ships Mage-VL 4B on Hugging Face as a streaming video-language model

multimodalart · x · 2026-07-27

Microsoft has released Mage-VL 4B on Hugging Face, describing it as an efficient codec-native streaming VLM.

The model can describe live events as they happen and can be prompted to focus on a specific moment, such as the instant a train arrives or when a goal is scored. A Spaces demo is available for testing.

Related event: Microsoft unveils Mage-VL, a 4B streaming multimodal model(5 posts)→

Original post →

More from Multimodal

Multimodal channel →