Parameters
4B
Context Length
262K
Modality
Text
Architecture
Dense
KV Cache Rate
144 KB/tok
License
Apache 2.0
Release Date
30 Oct 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Self-hosted only
Precision
Context
Workload Memory: 5.49 GB
INT4 · 8K context
Base Weights: 1.86 GB
KV Cache: 1.13 GB
Runtime & Graphs: 2.50 GB
1x NVIDIA L4 24GB
5.5 / 21.6 GB usable
1x RTX 3060 12GB
5.5 / 10.8 GB usable
Apple M4 24GB
5.5 / 18.0 GB usable
Determine requirements for your specific workload:
Open in VRAM Calculator
Checkpoint format
FP16
CUDA
>= 11.8
PyTorch
>= 2.1
Transformers
>= 4.57.3
| Architecture | Precision | Attention kernel |
|---|---|---|
Blackwell, Hopper | Native FP16 | FlashAttention-3 |
Ada Lovelace, Ampere | Native FP16 | FlashAttention-2 |
Additional required packages
Install alongside standard runtime libraries:
pip install flash_attn
No evaluation benchmarks for Mage-VL Cowork available.
Overall Rank
-
Coding Rank
-
Mage-VL Cowork is a multimodal vision-language model developed by Microsoft tailored for collaborative workplace interfaces and GUI interaction. It leverages an advanced one-vision visual backbone to parse complex user interface workflows.
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
32
Key-Value Heads
8
Attention Head Dimension
128
Position Embedding
ROPE
RoPE Theta
5,000,000
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Base Precision
FP16
Hidden Dimension Size
2,560
Number of Layers
36
FFN Intermediate Size (Dense)
9,728
Tokenizer
Vocabulary Size
151,936
The Mage-VL model family developed by Microsoft.