• 🔥[Feb. 2026] We release the GUI-Owl 1.5 model family on HuggingFace.

• Open source GUI-Owl 1.5 model weights
• Open source GUI-Owl 1.5 model cookbook
• Deploy Mobile-Agent-v3.5 on your own device
• Open source evaluation code on benchmarks

GUI-Owl 1.5 is the next-generation native GUI agent model family built on Qwen3-VL. It supports multi-platform GUI automation across desktops, mobile devices, browsers, and more. Powered by a scalable hybrid data flywheel, unified agent capability enhancement, and multi-platform environment RL (MRPO), GUI-Owl 1.5 offers a full spectrum of models:

Key highlights:

• 🏆 State-of-the-art among multi-platform GUI models on OSWorld-Verified, AndroidWorld, Mobile-World, WindowsAA, ScreenSpot-v2, ScreenSpot-Pro, and more.
• 🔧 Tool & MCP calling: Native support for external tool invocation and MCP server coordination, achieving top performance on OSWorld-MCP and Mobile-World.
• 🧠 Long-horizon memory: Built-in memory capability without external workflow orchestration, leading all native agent models on MemGUI-Bench.
• 🤝 Multi-agent ready: Serves both as a standalone end-to-end agent and as specialized roles (planner, executor, verifier, notetaker) within the Mobile-Agent-v3.5 framework.
• ⚡ Instruct & Thinking variants: Smaller instruct models for fast inference and edge deployment; larger thinking models for complex tasks requiring planning and reflection.

❗ At present, only Android OS support tool debugging. Other systems, such as iOS, do not support the use of Mobile-Agent for the time being.

pip install qwen_agent
pip install qwen_vl_utils
pip install numpy

1. Download the Android Debug Bridge.
2. Turn on ADB debugging on your Android phone (enable Developer Options first). For HyperOS, also enable "USB Debugging (Security Settings)".
3. Connect your phone to the computer with a data cable and select "Transfer files".
4. Test your ADB environment: adb devices. If connected devices are displayed, you're ready.
5. On Mac/Linux, ensure ADB permissions: sudo chmod +x /path/to/adb
6. On Windows, use: xx\xx\adb.exe

1. Download the ADB Keyboard APK.
2. Install on your mobile device.
3. Switch the default input method to "ADB Keyboard" in system settings.

cd Mobile-Agent-v3.5/mobile_use
python run_gui_owl_1_5_for_mobile.py \
    --adb_path "Your ADB path" \
    --api_key "Your api key of vllm service" \
    --base_url "Your base url of vllm service" \
    --model "Your model name of vllm service" \
    --instruction "The instruction you want Mobile-Agent-v3.5 to complete" \
    --add_info "Some supplementary knowledge, can also be empty"

1. GUI-Owl 1.5 outputs relative coordinates (0–1000) by default.

pip install pyautogui
pip install pyperclip

cd Mobile-Agent-v3.5/computer_use
python run_gui_owl_1_5_for_pc.py \
    --api_key "Your api key of vllm service" \
    --base_url "Your base url of vllm service" \
    --model "Your model name of vllm service" \
    --instruction "The instruction you want Mobile-Agent-v3.5 to complete" \
    --add_info "Some supplementary knowledge, can also be empty"

1. GUI-Owl 1.5 outputs relative coordinates (0–1000) by default.

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium

# Model API keys (required)
export API_KEY="sk-xxx"          # Agent model API
export OMNI_API_KEY=""
export EVAL_API_KEY="sk-xxx"               # Evaluation model API

cd cd Mobile-Agent-v3.5/browser_use
python run_gui_owl_1_5_for_web.py \
  --task "Search for 'Tongyi Lab'" \
  --web "https://bing.com" \
  --base_url "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions" \
  --model "claude-sonnet-4-5-20250929" \
  --output_dir results/custom \
  --image_type base64 \
  --headless \
  --use_css_som

Please refer to Link.

1. Please follow the official code repository to install the Android emulator and necessary dependencies.

2. Install the dependencies.

cd Mobile-Agent-v3.5/android_world
pip install -r requirements.txt

3. Fill in your vllm service information in the run_guiowl15.sh or run_ma35.sh script, including api_key, base_url, and model.

4. Run the evaluation.

sh run_guiowl15.sh
sh run_ma35.sh

1. Please download the images and annotations for the grounding benchmarks from their official repository.

2. Install the dependencies required by the qwen model.

pip install qwen_agent
pip install qwen_vl_utils

3. Fill in your information in the run_grounding.sh, including MODEL_PATH, DS_PATH, SAVE_PATH and EVAL_TYPE.

4. Run the evaluation.

cd Mobile-Agent-v3.5/grounding_and_kb
sh run_grounding.sh

1. OSWorld-MCP: Please follow the official code repository to run the evalutation
2. MobileWorld: Please follow the official code repository to run the evalutation

1. Download the images and annotations for the Knowledge Bench from the official repository. Follow the official instructions to draw the GUI actions, and save the annotated images to DS_PATH/AnnotateImage.

2. Install the dependencies required by the qwen model.

pip install qwen_agent
pip install qwen_vl_utils

3. Fill in your information in the run_gui_kb.sh, including MODEL_PATH, DS_PATH, SAVE_PATH and EVAL_TYPE.

4. Run the evaluation.

cd Mobile-Agent-v3.5/grounding_and_kb
sh run_gui_kb.sh

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium

# Model API keys (required)
export API_KEY="sk-xxx"          # Agent model API
export OMNI_API_KEY=""
export EVAL_API_KEY="sk-xxx"               # Evaluation model API

• WebArena/VisualWebArena: Start the corresponding environment services in advance (configure ports per the official docs).
• Port mapping: Ports are preset in the code (see visualwebarena_url_map / webarena_url_map). To change them, modify BASE_URL and the port numbers.
• Task data: Ensure data/merged_test_raw.json exists (includes task definitions, login requirements, initial screenshots, etc.).

# WebVoyager
cd Mobile-Agent-v3.5/web_benchmark
python main_for_eval.py \
  --task "Find out which four teams the NFC North contains in the NFL on ESPN." \
  --web https://www.espn.com/ \
  --output_dir results/WebVoyager \
  --image_type file \
  --task_id "validation_WebVoyager__ESPN--41" \
  --use_css_som \
  --headless

Please refer to the technical report for detailed results on ScreenSpot-v2, ScreenSpot-Pro, OSWorld-G, MMBench-GUI, and more.

from transformers import Qwen3VLForConditionalGeneration, AutoProcessor

model_name = "mPLUG/GUI-Owl-1.5-8B-Instruct"
model = Qwen3VLForConditionalGeneration.from_pretrained(
    model_name, torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained(model_name)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "screenshot.png"},
            {"type": "text", "text": "Click on the search bar."},
        ],
    }
]

inputs = processor.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_dict=True,
    return_tensors="pt"
)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
    out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
    generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)

If you find GUI-Owl 1.5 useful in your research, please cite:

@article{xu2026mobile,
  title={Mobile-Agent-v3. 5: Multi-platform Fundamental GUI Agents},
  author={Xu, Haiyang and Zhang, Xi and Liu, Haowei and Wang, Junyang and Zhu, Zhaozai and Zhou, Shengjie and Hu, Xuhao and Gao, Feiyu and Cao, Junjie and Wang, Zihua and others},
  journal={arXiv preprint arXiv:2602.16855},
  year={2026}
}

GUI-Owl 1.5 is built upon Qwen3-VL. We thank the Qwen team for their excellent open-source foundation models.