furiosa-ai/Qwen3-VL-4B-Thinking

VISIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Qwen3-VL-4B-Thinking by Qwen is a 4-billion-parameter dense vision-language model from the Qwen3-VL series, designed for multimodal understanding tasks. It integrates a vision encoder with a transformer decoder, supporting image and video input alongside text. This "Thinking" edition explicitly generates a chain-of-thought before its final answer, enhancing reasoning for tasks like OCR, document analysis, and tool calling.

Loading preview...

Overview

Qwen3-VL-4B-Thinking is a 4-billion-parameter dense vision-language model developed by Qwen. It is part of the Qwen3-VL series, combining a vision encoder with a dense transformer decoder. This model utilizes Interleaved-MRoPE positional embeddings and DeepStack multi-level feature fusion to process images and videos alongside text inputs. A key differentiator is its "Thinking" edition, which generates an explicit chain-of-thought before providing a final answer, aiding in complex reasoning tasks.

Key Capabilities

  • Multimodal Understanding: Handles visual understanding tasks including OCR, document and chart analysis, spatial reasoning, and video comprehension.
  • Chain-of-Thought Reasoning: Emits a detailed chain-of-thought, parsed by Furiosa-LLM's qwen3 reasoning parser, to separate reasoning steps from the final answer.
  • Tool Calling: Natively supports tool (function) calling, parsed via the hermes tool-call parser, enabling integration with external functionalities.
  • Flexible Input: Accepts OpenAI-style multimodal chat messages with image_url content parts, supporting remote URLs, base64 data, and local file paths.

Use Cases

  • Complex Visual Analysis: Ideal for applications requiring detailed interpretation of visual data, such as analyzing charts, documents, or performing OCR.
  • Reasoning-Intensive Tasks: Suitable for scenarios where an explicit reasoning process is beneficial for transparency or debugging.
  • Automated Workflows: Can be integrated into systems requiring tool use or function calling based on multimodal inputs.

This specific repository provides the model with a Furiosa Executable Bundle (FXB) for optimized execution on FuriosaAI RNGD hardware using Furiosa-LLM, while also being compatible with other frameworks like vLLM and Transformers.

Finetunes
1 models