Mage-VL-Cowork
Mage-VL-Cowork is an open-weight vision-language model from Microsoft, released in October 2026. It is built for computer use: it reads screenshots and chooses between mouse and keyboard actions and shell commands.
- Developer
- Microsoft
- Released
- October 2026
- Total parameters
- 4.7B
- Architecture
- Dense
- Context window
- 32K
- Input
- Text, image
- License
- Apache 2.0
- Commercial use
- Yes
- Official weights, GB
- 9.5 GB
- Hugging Face repository
- huggingface.co
- Checked on
Mage-VL-Cowork is a dense model with 4.7B parameters (Microsoft calls it a 4B model). Microsoft lists a context window of 32,768 tokens for training and inference; the architecture supports up to 262,144 tokens. The model accepts text and image input. The weights are released under Apache 2.0, which allows commercial use. The official weights on Hugging Face take about 9.5 GB in BF16. Running the model needs at least that much memory across GPUs and system RAM, plus room for the context cache; quantized versions need less.
- Total parameters, billions
- 4.7B
- Context window, tokens
- 32.77K
- License type
- Apache 2.0
- License conditions
- Apache 2.0: keep the license and notices; no limits on field of use or number of users.
- Weights precision
- BF16
- License text
- apache.org
- Sources
Page URL Hugging Face model card https://huggingface.co/microsoft/Mage-VL-Cowork/raw/main/README.md Hugging Face LICENSE https://huggingface.co/microsoft/Mage-VL-Cowork/raw/main/LICENSE Hugging Face weights index https://huggingface.co/microsoft/Mage-VL-Cowork/raw/main/model.safetensors.index.json Hugging Face API https://huggingface.co/api/models/microsoft/Mage-VL-Cowork
Questions
What license does Mage-VL-Cowork use?
Can I use Mage-VL-Cowork commercially?
How much memory does Mage-VL-Cowork need?
How many parameters does Mage-VL-Cowork have?
Does Mage-VL-Cowork accept images?
Sources
Every figure comes from the developer’s model card, license text and official announcements, and from the model’s Hugging Face repository metadata. Each model page lists its sources and the date they were checked; the register is rechecked every week.