# Qwen3 VL 235B A22B Instruct

Qwen3 VL 235B A22B Instruct is a vision-language model that processes interleaved text, images, and video within a native context of 262.1K tokens, purpose-built for instruction-following tasks that span documents, visual reasoning, and screen-based agent workflows.

- **Model ID:** `alibaba/qwen3-vl-instruct`
- **Type:** chat
- **Providers:** deepinfra
- **Context window:** 262,144
- **Maximum output tokens:** 262,144
- **Pricing:** $0.2/1M input tokens, $0.88/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/qwen3-vl-instruct

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'alibaba/qwen3-vl-instruct',
  prompt: 'Why is the sky blue?'
})
```

## About

Qwen3 VL 235B A22B Instruct is the general-purpose variant in the Qwen3-VL model family, built on a mixture-of-experts (MoE) architecture with 235 billion total parameters and approximately 22 billion active per token. Its context window of 262.1K tokens accommodates interleaved sequences of text, images, and video frames, making it practical for reasoning across large multimodal documents without segmenting input.

Three architectural innovations distinguish Qwen3-VL from prior generations. Enhanced interleaved Multimodal Rotary Position Embedding (MRoPE) improves spatial and temporal modeling across visual inputs, giving Qwen3 VL 235B A22B Instruct a stronger sense of object positions within images and event ordering within video. DeepStack integration fuses multi-level Vision Transformer (ViT) features from shallow, middle, and deep layers to tighten alignment between visual tokens and language tokens, improving grounding precision. Text-based temporal alignment for video replaces the prior T-RoPE approach with explicit textual timestamp grounding, enabling more reliable event localization within long video sequences.

Qwen3 VL 235B A22B Instruct extends its vision capabilities to agentic scenarios: it can parse GUI screenshots, understand layout and interactive elements, and plan actions for PC or mobile automation workflows. Optical character recognition (OCR) covers 32 languages and handles challenging conditions including low light, blurred text, and tilted documents. On standard multimodal benchmarks including MMMU and visual-math evaluations (MathVista, MathVision), Qwen3 VL 235B A22B Instruct reports competitive results against other frontier vision-language models.

## What to consider

For applications that process video or multi-image inputs, confirm that your selected provider's serving infrastructure supports large multimodal payloads at your target throughput before routing production traffic.

## When to use

### Best For

- Document intelligence pipelines that require OCR across 32 languages under varied image quality conditions
- GUI automation and screen-reading agents that interpret application screenshots to plan and execute UI actions
- Long video comprehension tasks where precise event localization and temporal reasoning over extended sequences are required
- Multi-image analysis workflows, comparing product photographs, reviewing multiple chart pages, or cross-referencing figures across a document
- Spatial reasoning tasks that require accurate 2D or 3D grounding of objects within images

### Consider Alternatives When

- If your task requires extended step-by-step reasoning traces over visual inputs, consider Qwen3-VL-Thinking for STEM and compositional visual reasoning
- If your workload is text-only and doesn't use vision, a text-only model will provide lower cost and faster throughput
- Latency is critical and your task is basic instruction following without complex visual analysis

## Best for

- **Multilingual document intelligence:** Pipelines that require OCR across 32 languages under varied image quality conditions
- **GUI automation and screen reading:** Agents that interpret application screenshots to plan and execute UI actions
- **Long video comprehension:** Tasks that need precise event localization and temporal reasoning over extended sequences
- **Multi-image analysis:** Comparing product photographs, reviewing multiple chart pages, or cross-referencing figures across a document
- **Spatial grounding:** Reasoning tasks that require accurate 2D or 3D grounding of objects within images

## Consider alternatives

- **Extended visual reasoning traces:** Consider Qwen3-VL-Thinking when STEM and compositional visual reasoning need step-by-step traces
- **Text-only workloads:** A text-only model will provide lower cost and faster throughput when vision isn't used
- **Latency-critical basic tasks:** Simple instruction following without complex visual analysis doesn't need this model's scale

## Frequently asked questions

### What modalities does Qwen3 VL 235B A22B Instruct accept?

The model accepts interleaved sequences of text, images, and video frames within a single context window of up to 262.1K tokens.

### How does DeepStack improve vision-language alignment?

DeepStack fuses feature maps from multiple depth levels of the Vision Transformer, shallow layers capture low-level detail while deeper layers encode abstract semantics. Combining these gives the language model richer visual grounding information than single-layer ViT representations.

### What is the difference between MRoPE and standard positional encoding for video?

Enhanced interleaved MRoPE assigns distinct positional axes to the temporal (time), height, and width dimensions of video inputs, giving the model an explicit spatial-temporal coordinate system. This improves its ability to reason about where and when events occur within a video sequence.

### Can this model perform GUI automation tasks?

Yes. The model is trained to parse GUI screenshots, identify interactive elements (buttons, forms, navigation), and plan multi-step actions for PC or mobile application automation.

### What OCR languages and conditions are supported?

The model covers OCR in 32 languages and has been evaluated for robustness in low-light, blurred, and tilted-text conditions.

### How does Qwen3 VL 235B A22B Instruct differ from Qwen3-VL-Thinking?

The Instruct variant is designed for direct instruction following and is generally faster and more cost-effective. The Thinking variant adds extended step-by-step reasoning traces optimized for complex STEM and compositional visual reasoning problems.

### Is the context window of 262.1K tokens shared across text, image, and video tokens together?

Yes. The limit of 262.1K tokens applies to the combined sequence of text tokens and visual tokens (image patches, video frames encoded as tokens) in an interleaved context.

## Links

- [Model page](https://vercel.com/ai-gateway/models/qwen3-vl-instruct)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://deepinfra.com/Qwen/Qwen3-VL-235B-A22B-Instruct)
- [Provider pricing](https://deepinfra.com/pricing)