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

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent 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 is a 4-billion-parameter dense vision-language model from the Qwen3-VL series, developed by Qwen. It integrates a vision encoder with a dense transformer decoder, utilizing Interleaved-MRoPE positional embeddings and DeepStack multi-level feature fusion. This "Thinking" edition is distinguished by its explicit chain-of-thought generation before providing a final answer, making it suitable for complex visual understanding tasks, OCR, document analysis, spatial reasoning, video comprehension, and native tool calling, especially when detailed reasoning steps are required.

Loading preview...

Overview

Qwen3-VL-4B-Thinking is a 4-billion-parameter dense vision-language model from the Qwen3-VL series, developed by Qwen. It combines a vision encoder with a dense transformer decoder, featuring Interleaved-MRoPE positional embeddings and DeepStack multi-level feature fusion. This specific "Thinking" edition is designed to emit an explicit chain-of-thought before its final answer, enhancing transparency and interpretability in its responses. The model is optimized for deployment on FuriosaAI RNGD hardware using Furiosa-LLM, running with a tensor-parallel size of 8 PEs per card.

Key Capabilities

  • Vision-Language Understanding: Processes both images and text inputs, supporting OpenAI-style multimodal chat messages with image_url content parts.
  • Reasoning: Generates a detailed chain-of-thought, which Furiosa-LLM can parse separately from the final answer, aiding in complex problem-solving.
  • Tool Calling: Supports native tool (function) calling via the hermes tool-call parser, enabling interaction with external functions.
  • Broad Visual Tasks: Capable of handling OCR, document and chart analysis, spatial reasoning, and video comprehension.

Good For

  • Applications requiring detailed, step-by-step reasoning for visual and textual queries.
  • Integrating vision-language capabilities with external tools or functions.
  • Deployments on FuriosaAI RNGD hardware seeking optimized performance for Qwen3-VL models.
  • Tasks involving complex visual understanding where transparency of thought process is beneficial.

Finetunes
1 models