Qwen: Qwen3 VL 235B A22B Instruct
Model Type
Open Weight Model
235B parameters
Recommended Use Cases
Text Generation
Try Qwen3 VL 235B A22B Instruct
Qwen3-VL-235B-A22B-Instruct is Alibaba's flagship open-weight vision-language model with 235B total parameters and 22B active per token, delivering state-of-the-art multimodal understanding for production deployment.
Overview
Qwen3-VL-235B-A22B-Instruct is the production-optimized flagship of the Qwen3-VL family. It rivals proprietary models like Gemini 2.5 Pro on perception benchmarks while remaining fully open-weight under Apache 2.0. The model excels at visual agent tasks, visual coding, and complex document understanding.
Key Features
- Flagship MoE: 235B total, 22B active per token
- Visual agent: Operates PC/mobile GUIs autonomously
- Visual coding: Generates Draw.io/HTML/CSS/JS from images and videos
- 3D grounding: Advanced spatial reasoning for embodied AI
- 1M context: 256K native, expandable to 1M tokens
- 32-language OCR: Robust in low light, blur, and tilt
Technical Specifications
| Specification | Value |
|---|---|
| Total Parameters | 235B |
| Active Parameters | 22B |
| Architecture | MoE transformer with MLA |
| Vision Encoder | DeepStack multi-level ViT fusion |
| Context Length | 256K tokens (expandable to 1M) |
| Release Date | September 2025 |
Architecture Innovations
- Interleaved-MRoPE: Positional embeddings for long-horizon video
- DeepStack: Multi-level ViT feature fusion for fine-grained alignment
- Text-Timestamp Alignment: Precise video temporal modeling
When to Use Qwen3-VL-235B-A22B-Instruct
Choose this model when you need:
- Best-in-class visual understanding
- Complex document and chart parsing
- GUI automation and visual agents
- Production multimodal applications
- Fast responses at flagship quality
Consider alternatives when you need:
- Maximum reasoning depth (use Thinking variant)
- Lower deployment costs (use 30B-A3B)
- Edge deployment (use 8B models)
Availability
- Open Weights: Hugging Face (Qwen/Qwen3-VL-235B-A22B-Instruct)
- API: OpenRouter, DeepInfra, Novita
- Local: vLLM with tensor parallelism (~471GB weights)
Role in Series
Qwen3-VL models by capability:
- Qwen3-VL-8B: Edge deployment
- Qwen3-VL-30B-A3B: Efficient MoE
- Qwen3-VL-235B-A22B-Instruct: Flagship production (this model)
- Qwen3-VL-235B-A22B-Thinking: Flagship reasoning