Skip to content
Computer vision

Mage-VL-Cowork

4B-агент, который сам выбирает: кликнуть или написать код

2 min readupdated

Microsoft Mage Team представила Mage-VL-Cowork — компактную 4B vision-language policy для hybrid computer-use. Один чекпоинт объединяет работу через скриншоты и GUI (computer_use) с выполнением кода и команд в терминале (bash), без внешнего оркестратора. Архитектура llava_onevision2: Mage-ViT + Qwen3-4B.

Ключевые особенности

  • Два канала в одной модели: turn-level routing между GUI и терминалом — для визуальных задач и структурированных правок файлов.
  • 66.0% на OSWorld-Verified при бюджете 50 шагов — лучший результат среди 4B-моделей; 87.5% на OS-задачах и 62.7% на multi-application workflows.
  • Пятиэтапный curriculum: сначала отдельные навыки GUI и кода, затем Hybrid SFT и online RL; прирост +20.44 п.п. по сравнению со стартовым чекпоинтом.
  • Mage-CUA-Forge: синтез 293 806 задач по 6 915 atomic skills; повторяющаяся стоимость генерации снижена на 82.6%.
  • Асимметричная память: 3 последних скриншота, 4 полных ответа и свёрнутая история Action:; контекст 32 768 токенов.

Пример

Сценарий из карточки: «В таблице на рабочем столе пересчитай столбец и проверь итог». Агент открывает файл через GUI, для массовых правок переключается на bash с pandas/openpyxl, а затем возвращается к экрану для визуальной проверки результата.

ПодробнееOSWorld-Verified: 66.0% (среднее двух прогонов; 65.33% и 66.75%).
Online GRPO даёт +10.16 п.п. к Hybrid SFT и сокращает число вызовов модели на 13.0%, output tokens — на 25.4%.
Веса: 4B, 9.48 ГБ в bfloat16 в двух safetensors shards; язык — английский; лицензия Apache-2.0.
Примеры
Discuss Telegram 232 5.9K

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.8 GB · ready files10 GB
  • FP8 / INT8recommendedweights 4.4 GB5.4 GB
  • 4-bitweights 2.5 GB3.2 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2080 Ti

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # microsoft/Mage-VL-Cowork
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download microsoft/Mage-VL-Cowork --local-dir ~/models/Mage-VL-Cowork
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Mage-VL-Cowork
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Mage-VL (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 4All on Hugging Face

About the model

microsoft/Mage-VL-Cowork
Author
microsoft
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

5.9K232

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Computer vision

Mage Hand

Mage Hand
Computer vision

d1-2B

d1-2B