Skip to content
Local AI Wizard
Your machine
Theme

Mage-VL

Microsoft · microsoft/Mage-VL

Weights license: Apache 2.0 · code and components may use other terms

Reported by the publisher46 resultsMicrosoft's model card · read 2026-10-25 · not ranked · see them
Memory needed~4 GBQ4 weights + working room, before the machine's reserve
Parameters4.7 BHub safetensors total
Repository updated2026-08-10A commit is not a release date

Quality

No comparable score

We track no public, comparable benchmark that covers this model yet. No score is inferred.

Reported by the publisher

46 results · Microsoft's model card · read 2026-10-25
DocVQA-val
95.14Document understanding
InfoVQA-val
80.33Document understanding
AI2D w/ Mask
83.16Document understanding
ChartQA
84.88Document understanding
OCRBench
81.80Document understanding
MultiDocVQA-val
87.46Document understanding
ChartQAPro
32.57Document understanding
TextVQA-val
77.28Document understanding
Every reported result46
CC-OCR Doc
32.25Document understanding
MMBench-EN-dev
84.02General VQA
MMBench-CN-dev
82.04General VQA
MMStar
67.32General VQA
MME-Perception
1,709.54General VQA
SeedBench (All)
76.78General VQA
CV-Bench
87.79General VQA
MME-RealWorld
66.52General VQA
CV-Bench-2D
82.13Spatial intelligence
CV-Bench-3D
94.75Spatial intelligence
BLINK
65.11Spatial intelligence
EmbSpatial
82.67Spatial intelligence
CrossPoint
80.00Spatial intelligence
CRPE-Relation
76.12Spatial intelligence
SAT
67.33Spatial intelligence
MV-Bench
65.1Video QA
NextQA
83.1Video QA
VideoMME
64.0Video QA
LongVideoBench
61.3Video QA
LVBench
41.8Video QA
MLVU-dev
68.7Video QA
VideoEval-Pro
45.2Video QA
Timelens-Charades
50.7Temporal grounding
Timelens-ActivityNet
45.4Temporal grounding
Timelens-QVHighlight
57.4Temporal grounding
VSI-Bench
64.3Spatial reasoning
Ref-DAVIS17
25.83Tracking (J&F)
MeViS-ValidU
22.55Tracking (J&F)
ReasonVOS
17.76Tracking (J&F)
Ref-YT-VOS
25.57Tracking (J&F)
TriggerAcc
79.21
TimVal
55.54
ROC-AUC
83.14
PR-AUC
9.30
RT-Avg
79.841 fps
BT-Avg
48.151 fps
Overall
64.001 fps

Values as printed on the card, with the publisher's own settings (card section: 📊 Performance); column “Mage-VL-4B”. Not comparable across publishers and not used in any ranking. Model card

Runs on

Fast memory holds it 118

The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.

Show all 118

Within memory, outside the fast part 1

Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.

Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Not a measured fit or speed.

Consider it for

Microsoft released Mage-VL on 2026-07-26: a vision-language model that answers about images, with 4.7B parameters on the Hugging Face count and the Apache 2.0 license.

Check first

License: commercial use allowed. At Q4 the weights load about 3.5 GB before each machine's reserve; image resolution, video frames and context add memory, and the vision encoder needs support in the runtime you pick.

Measured speed

No published measurement for this checkpoint

Tokens per second are shown only when a source publishes them with hardware, build and setup. They are never estimated.

How to run

Runtime compatibility and tokens per second have not been measured for this entry. Published metadata and a memory estimate do not prove a successful local run.

Library in source: transformers

Documented ways to run it

Copied as printed in the linked source (read 2026-10-18); not run by this site.

Mage-VL · model card · Offline inference
# image
python inference.py --mode offline --image examples/dog.jpg \
  --question "Describe this image in detail."
SGLang · model card · Online inference
sudo apt-get update && sudo apt-get install -y protobuf-compiler
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs \
  | sh -s -- -y --profile minimal --default-toolchain 1.90.0
source "$HOME/.cargo/env"

git clone -b feat/mage-vl https://github.com/kcz358/sglang
cd sglang
pip install -e 'python[all]'
python -m sglang.launch_server \
  --model-path microsoft/Mage-VL \
  --trust-remote-code

Choose my computer

Technical details

What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.

Checkpoint / source
microsoft/Mage-VL
Task
image-text-to-text
Architecture
MageVLForConditionalGeneration
Library in source
transformers
Weight license
Apache 2.0microsoft/Mage-VL model card (Hugging Face) · read 2026-10-18source
Commercial use
Allowed Apache License 2.0 text (SPDX) · read 2026-10-18license text
Released
Mage README news, '2026-07-26: Mage-VL released on Hugging Face' (microsoft/Mage GitHub) · read 2026-10-18source
Every stored field11 fields
Revision checked
d88b153285f1633a61b2f693c59c8576693af185
Architecture
MageVLForConditionalGeneration
Library
transformers
Hub task tag
image-text-to-text
Formats
transformers, safetensors
Access
Open download
Publisher kind
Official
Repository created
2026-07-25creation is not the model's release
Repository updated
2026-08-10a commit is not a release date
Hub downloads (30 days) · likes
8,197 · 440this repository only · read 2026-10-25
Catalog state
Hub config or weights found; no inventory benchmark link

Sources

Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.

Curated facts checked: 2026-10-18. The license and release date cite the publisher's sources; parameters, files and task come from the Hugging Face record, and memory is a labelled estimate from its parameter count. A value no source states stays empty; a repository commit is not a model release date.