Zen Designer is Zen LM's flagship vision-language model: 235B total parameters with 22B active via Mixture of Experts (MoE). It delivers comprehensive visual understanding — images, video, documents, charts, and GUIs — combined with state-of-the-art language capabilities.

Two variants are available:

instruct: optimized for direct visual question answering and task completion
thinking: extended chain-of-thought reasoning for complex visual analysis

Model Family

Model	Type	Context	Description
zen-designer-235b-a22b-instruct	Instruct	256K	Direct VLM tasks
zen-designer-235b-a22b-thinking	Thinking	256K	Extended reasoning

Model Specifications

Attribute	Value
Total Parameters	235B
Active Parameters	22B (MoE)
Architecture	Dense vision encoder + MoE language decoder
Context Window	256K tokens (expandable to 1M)
Languages	100+ (OCR in 32 scripts)
Video Support	Long-form video understanding
License	Apache 2.0

Key Capabilities

Visual Understanding

Detailed image analysis and description
Video understanding with temporal reasoning
Document parsing: PDFs, invoices, forms, tables, charts
OCR across 32 languages and writing systems

Spatial and Structural Reasoning

2D and 3D spatial grounding
Bounding box prediction
GUI element recognition and navigation
Code generation from UI screenshots (HTML/CSS/JS)

Agentic Vision

Web navigation with visual context
GUI interaction and automation
Computer use workflows
Diagram comprehension (Draw.io, flowcharts, schematics)

Quick Start

Install

pip install transformers torch accelerate

Image Understanding

from transformers import AutoModelForVision2Seq, AutoProcessor
import torch
from PIL import Image

model_name = "zenlm/zen-designer-235b-a22b-instruct"

processor = AutoProcessor.from_pretrained(model_name)
model = AutoModelForVision2Seq.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

image = Image.open("document.png")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Extract all text and data from this document in structured format."}
        ]
    }
]

text = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=text, images=image, return_tensors="pt").to(model.device)

with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=2048)

response = processor.decode(generated_ids[0], skip_special_tokens=True)
print(response)

Video Understanding

from transformers import AutoModelForVision2Seq, AutoProcessor
import torch

model_name = "zenlm/zen-designer-235b-a22b-instruct"

processor = AutoProcessor.from_pretrained(model_name)
model = AutoModelForVision2Seq.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "video", "video": "path/to/video.mp4", "fps": 2},
            {"type": "text", "text": "Describe what happens in this video, step by step."}
        ]
    }
]

text = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=text, return_tensors="pt").to(model.device)

with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=1024)

response = processor.decode(generated_ids[0], skip_special_tokens=True)
print(response)

Extended Thinking (Complex Tasks)

from transformers import AutoModelForVision2Seq, AutoProcessor
import torch
from PIL import Image

model_name = "zenlm/zen-designer-235b-a22b-thinking"

processor = AutoProcessor.from_pretrained(model_name)
model = AutoModelForVision2Seq.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

image = Image.open("math_problem.png")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image"},
            {"type": "text", "text": "Solve this problem step by step, showing your reasoning."}
        ]
    }
]

text = processor.apply_chat_template(messages, add_generation_prompt=True)
inputs = processor(text=text, images=image, return_tensors="pt").to(model.device)

with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=8192)

response = processor.decode(generated_ids[0], skip_special_tokens=True)
print(response)

Performance Benchmarks

Benchmark	Score	Category
DocVQA	State-of-the-art	Document understanding
ChartQA	State-of-the-art	Chart reasoning
MMBench	Leading	Multimodal understanding
OCRBench	Leading	OCR accuracy
Video-MME	Leading	Video reasoning
MMMU	Competitive	Multi-discipline QA

Hardware Requirements

Setup	VRAM	Notes
Minimum (INT4)	4x 24GB	Quantized inference
Recommended (BF16)	4x 80GB	Full precision
Optimal	8x 80GB	Maximum throughput

Deployment

# vLLM (recommended for production)
vllm serve zenlm/zen-designer-235b-a22b-instruct \
    --tensor-parallel-size 4 \
    --max-model-len 65536

# SGLang
python -m sglang.launch_server \
    --model-path zenlm/zen-designer-235b-a22b-instruct \
    --tp-size 4

License

Apache 2.0

Citation

@misc{zenlm2025zen-designer,
    title={Zen Designer: 235B Vision-Language Model by Zen LM},
    author={Hanzo AI and Zoo Labs Foundation},
    year={2025},
    publisher={HuggingFace},
    howpublished={\url{https://huggingface.co/zenlm/zen-designer-235b-a22b-instruct}}
}

Zen LM by Hanzo AI - Clarity Through Intelligence
zenlm.org | HuggingFace | GitHub