furiosa-ai/Qwen3-VL-32B-Instruct

VISIONPricing:Input $0.416 / Output $1.664Concurrent Unit Cost:2Model Size:33.4BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Qwen3-VL-32B-Instruct by Qwen is a 33.4 billion parameter dense vision-language model from the Qwen3-VL series, designed to process both images and text. It features Interleaved-MRoPE positional embeddings and DeepStack multi-level feature fusion, enabling it to handle visual understanding tasks like OCR, document and chart analysis, spatial reasoning, and video comprehension. This instruction-tuned model also natively supports tool calling, making it suitable for multimodal applications requiring visual and textual input processing.

Loading preview...

Overview

Qwen3-VL-32B-Instruct is a 32-billion-parameter dense vision-language model developed by Qwen, designed to process both image and text inputs. This instruction-tuned variant is optimized for multimodal chat messages and supports advanced features like tool calling. It utilizes Interleaved-MRoPE positional embeddings and DeepStack multi-level feature fusion for robust visual understanding.

Key Capabilities

  • Vision-Language Processing: Accepts OpenAI-style multimodal chat messages with image_url content parts alongside text, enabling comprehensive visual and textual understanding.
  • Tool Calling: Supports tool (function) calling through the hermes tool-call parser, allowing the model to interact with external functions.
  • Broad Visual Understanding: Capable of handling tasks such as OCR, document and chart analysis, spatial reasoning, and video comprehension.
  • Optimized for FuriosaAI Hardware: This specific repository provides a Furiosa Executable Bundle (FXB) for efficient deployment on FuriosaAI RNGD, running with a tensor-parallel size of 32 PEs across four RNGD cards.

Good for

  • Multimodal Applications: Ideal for use cases requiring the processing and understanding of both images and text simultaneously.
  • Automated Data Extraction: Suitable for tasks like OCR and analysis of documents and charts.
  • Interactive AI Systems: Its tool-calling capability makes it well-suited for building agents that can interact with external tools and APIs.
  • Deployment on FuriosaAI: Specifically configured for optimal performance and ease of deployment on FuriosaAI RNGD hardware.

Finetunes
1 models