Qwen3-VL-4B-Instruct

Qwen3-VL-4B-Instruct is a compact 4-billion-parameter multimodal model designed for efficient deployment on resource-constrained servers while retaining the full functionality of the Qwen3-VL series. Despite being half the size of the 8B version, the model preserves all key architectural innovations: Interleaved-MRoPE for video understanding, DeepStack for multi-level visual feature fusion, and Text-Timestamp Alignment for precise temporal localization. The seamless integration of text and visual modalities provides an understanding of multimodal context at a level comparable to pure-text LLMs.

In terms of performance, Qwen3-VL-4B-Instruct approaches the results of Qwen2.5-VL-7B, demonstrating that the reduction in model size was achieved without significant loss of quality. The model supports a native context of 256K tokens (expandable to 1M), enabling the processing of long documents, multi-hour videos, and complex multimodal dialogues. Advanced OCR capabilities with support for 32 languages and resilience to challenging shooting conditions make the 4B model a full-fledged solution for intelligent document processing tasks, despite its compact size.

Qwen3-VL-4B-Instruct represents an ideal solution for scenarios requiring a balance between performance and efficiency: deployment on consumer devices, the ability to process large volumes of visual content, fast response times for integration into real-time applications, and research projects. Furthermore, the open Apache 2.0 license allows for free commercial use of the model, making it accessible to a wide range of users, from startups to large enterprises.

Announce Date: 15.10.2025
Parameters: 4B
Context: 263K
Layers: 36
Attention Type: Full Attention
Developer: Qwen
Transformers Version: 4.57.0.dev0
License: Apache 2.0

Use our pre-built public endpoints for free to test inference and explore Qwen3-VL-4B-Instruct capabilities. You can obtain an API access token on the token management page after registration and verification.

An endpoint is a server with pre-installed software that automatically downloads neural network weights and launches an OpenAI-compatible API to handle incoming requests.

A private endpoint is created within the user's project, belongs exclusively to the user, and is billed per server configuration in accordance with the infrastructure pricing.

We recommend creating private endpoints when you need to:

• achieve maximum endpoint performance,
• access the full context window for processing large requests and conversations,
• ensure maximum security by processing data in a dedicated, isolated environment.

Contact our dedicated neural networks support team at support@immers.cloud or send your request to the sales department at sale@immers.cloud.