Happy Horse 1.0 is now on ModelsLab

Try Now
Skip to main content
Available now on ModelsLab · Language Model

Qwen: Qwen3 VL 235B A22B ThinkingThink Visually. Reason Deeply

Try Qwen: Qwen3 VL 235B A22B ThinkingAPI Documentation

Unlock Multimodal Intelligence

Visual Agent

Operate GUIs Autonomously

Recognizes elements, understands functions, invokes tools in PC/mobile interfaces.

Spatial Reasoning

Master 2D 3D Grounding

Judges positions, viewpoints, occlusions for spatial tasks and embodied AI.

Video Comprehension

Handle 1M Token Contexts

Processes hours-long videos with full recall and second-level indexing.

Examples

See what Qwen: Qwen3 VL 235B A22B Thinking can create

Copy any prompt below and try it yourself in the playground.

Diagram to Code

“Convert this flowchart image to Draw.io XML code. Ensure all nodes and connections match exactly. Output only the XML.”

Spatial Analysis

“Analyze this architectural blueprint: identify object positions, viewpoints, occlusions, and provide 3D grounding coordinates for key elements.”

Video Timeline

“From this 30-minute product demo video, extract second-level events: describe UI changes at 00:15, 02:30, and generate timeline-aligned text summary.”

STEM Reasoning

“Given this physics diagram image, solve the causal chain: compute forces, predict motion trajectory, explain step-by-step with evidence.”

For Developers

A few lines of code.
Vision reasoning. One call.

ModelsLab handles the infrastructure: fast inference, auto-scaling, and a developer-friendly API. No GPU management needed.

  • Serverless: scales to zero, scales to millions
  • Pay per token, no minimums
  • Python and JavaScript SDKs, plus REST API
API Documentation 
import requests


response = requests.post(
    "https://modelslab.com/api/v7/llm/chat/completions",
    json={
  "key": "YOUR_API_KEY",
  "prompt": "",
  "model_id": ""
}
)
print(response.json())

FAQ

Common questions about Qwen: Qwen3 VL 235B A22B Thinking

Read the docs

Qwen: Qwen3 VL 235B A22B Thinking is a MoE vision-language model with 235B total parameters, 22B active. It excels in multimodal reasoning for STEM, math, and visual tasks. Supports text, image, video inputs with 256K native context.

Access via OpenAI-compatible endpoints with base64 images or video URLs. Send multimodal messages for reasoning outputs. Handles visual coding, agents, long contexts seamlessly.

Features Thinking mode for step-by-step reasoning on complex visuals. Includes Interleaved-MRoPE for video and DeepStack for fine details. SOTA on perception, spatial, agent benchmarks.

Outputs at 56+ tokens/second, above average for size. MoE architecture activates 22B params efficiently. Balances speed and depth in Thinking/Non-Thinking modes.

Competes with top models like DeepSeek-R1, o1 in coding, math, vision tasks. Stronger visual agent, video understanding than prior VLMs. Ideal for document AI, UI automation.

Yes, supports hours-long videos with 1M expandable context. Provides timeline queries, full recall, dynamics comprehension. Uses second-level indexing for precision.

Ready to create?

Start generating with Qwen: Qwen3 VL 235B A22B Thinking on ModelsLab.

Try Qwen: Qwen3 VL 235B A22B ThinkingAPI Documentation