[Models](/models)/[Captioning](/models?category=captioning)/Qwen/Qwen3-VL-4B-Instruct

HFScene CaptioningApache-2.0

# Qwen3-VL-4B-Instruct

by Qwen

Best-in-class 4B vision-language model with 256K context and 32-language OCR

580Kdl/month

4.4Bparams

[HuggingFace](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`Qwen/Qwen3-VL-4B-Instruct`

## Deploy Qwen3-VL-4B-Instruct

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

Qwen3-VL-4B-Instruct is a dense 4.4B-parameter vision-language model with a three-module architecture: vision encoder, MLP-based vision-language merger, and LLM decoder. It supports 256K-1M context, 32-language OCR, native video temporal reasoning, and strong document understanding with 95.3% on DocVQA and 88.1% on OCRBench.

On Mixpeek, Qwen3-VL-4B powers scene captioning, visual question answering, and document understanding at the 4B parameter sweet spot, offering the best quality-to-cost ratio for pipelines that need both visual and text comprehension.

## Architecture

Dense transformer (36 layers, GQA 32/8) with 4.44B parameters. Three-module design: vision encoder, MLP vision-language merger, and LLM decoder. Interleaved-MRoPE for video temporal reasoning, DeepStack for multi-level ViT feature fusion, and Text-Timestamp Alignment for event localization.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so Qwen3-VL-4B-Instruct runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace\_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "asset-00412",
          // The model produces text, so it lands in payload. Give the
          // collection a text vector index and embed that text to make it
          // searchable rather than only filterable.
          payload: { extracted\_text: modelOutput, source\_key: "archive/2026/asset-00412" },
          vectors: { "multimodal-embedding": embeddingOfModelOutput },
        },
      \],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// universal\_extractor@v1 runs google/gemini-embedding-2
// (3072-d) over a bucket, with no inference of your own.

## Capabilities

*   256K-1M context window
*   32-language OCR and document understanding
*   Native video temporal reasoning with timestamp alignment
*   95.3% DocVQA, 88.1% OCRBench
*   Apache 2.0 license

## Use Cases on Mixpeek

Document understanding and extraction (invoices, forms, contracts)

Video scene captioning with temporal event localization

Multilingual OCR across diverse document types and languages

## Benchmarks

Dataset

Metric

Score

Source

DocVQA (test)

Accuracy

95.3%

Qwen, 2025: Qwen3-VL Technical Report

OCRBench

Score

88.1%

Qwen, 2025: Qwen3-VL Technical Report

MMBench-V1.1

Score

85.1%

Qwen, 2025: Qwen3-VL Technical Report

## Performance

Input SizeUp to 256K tokens (text + image patches)

GPU Latency~80ms / image (A100)

GPU Throughput~95 images/sec (A100, batch 8)

GPU Memory~9.5 GB

## Common Pipeline Companions

[

Qwen/Qwen3-Embedding-8B

Embed captions for semantic search

](/model/Qwen/Qwen3-Embedding-8B)[

Qwen/Qwen3-Reranker-8B

Rerank caption search results

](/model/Qwen/Qwen3-Reranker-8B)

## Explore on Mixpeek

[

More Captioning models

Compare alternatives in this category

](/models?category=captioning)[

Best Document AI Platforms

Hand-picked tools & platforms compared

](/curated-lists/best-document-ai-platforms)[

OCR & Document AI Internals

Deep-dive technical guide

](/guides/ocr-document-ai-internals)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

FrameworkHF

OrganizationQwen

FeatureScene Captioning

Outputtext

Modalitiesvideo, image

RetrieverSemantic Search

Parameters4.4B

LicenseApache-2.0

Downloads/mo580K

[

### Research Paper

Qwen3-VL Technical Report

arxiv.org](https://arxiv.org/abs/2511.21631)

### Build a pipeline with Qwen3-VL-4B-Instruct

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

### Alternative Models

[

Salesforce/blip2-opt-2.7b

Scene Captioning

](/model/Salesforce/blip2-opt-2.7b)[

microsoft/Florence-2-large

Scene Captioning

](/model/microsoft/Florence-2-large)[

google/paligemma2-3b-mix-448

Scene Captioning

](/model/google/paligemma2-3b-mix-448)[

Qwen/Qwen3-VL-8B-Instruct

Scene Captioning

](/model/Qwen/Qwen3-VL-8B-Instruct)