Qwen2.5-VL Technical Report

By Shuai Bai, Ke-qin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang and 17 more. arXiv.org, 2025

5,337 citations recorded by the source corpus.

We introduce Qwen2.5-VL, the latest flagship model of Qwen vision-language series, which demonstrates significant advancements in both foundational capabilities and innovative functionalities. Qwen2.5-VL achieves a major leap forward in understanding and interacting with the world through enhanced visual recognition, precise object localization, robust document parsing, and long-video comprehension. A standout feature of Qwen2.5-VL is its ability to localize objects using bounding boxes or points accurately. It provides robust structured data extraction from invoices, forms, and tables, as well as detailed analysis of charts, diagrams, and layouts. To handle complex inputs, Qwen2.5-VL introduces dynamic resolution processing and absolute time encoding, enabling it to process images of varying sizes and videos of extended durations (up to hours) with second-level event localization. This allows the model to natively perceive spatial scales and temporal dynamics without relying on traditional normalization techniques. By training a native dynamic-resolution Vision Transformer (ViT) from scratch and incorporating Window Attention, we reduce computational overhead while maintaining native resolution. As a result, Qwen2.5-VL excels not only in static image and document understanding but also as an interactive visual agent capable of reasoning, tool usage, and task execution in real-world scenarios such as operating computers and mobile devices. Qwen2.5-VL is available in three sizes, addressing diverse use cases from edge AI to high-performance computing. The flagship Qwen2.5-VL-72B model matches state-of-the-art models like GPT-4o and Claude 3.5 Sonnet, particularly excelling in document and diagram understanding. Additionally, Qwen2.5-VL maintains robust linguistic performance, preserving the core language competencies of the Qwen2.5 LLM.

Subfields: Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/f61cc9b5583c6295d5cd756ec0f34e4c003aab29

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Electrical and Electronic Engineering

1. Fei Huang

Citations in the corpus

• LVBench: An Extreme Long Video Understanding Benchmark (2024, neutral): “5-VL-72B [4], Gemini-2.”
• Video-R1: Reinforcing Video Reasoning in MLLMs (2025, neutral): “5-VL-72B [1] to generate COT rationales for the samples in Video-R1-260k.”
• InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency (2025, neutral): “5-VL-72B [5].”
• V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (2025, neutral): “The majority of the encoders used in MLLMs for VidQA are typically image encoders, which are applied independently per-frame for video inputs (Qwen Team et al., 2025; Zhang et al., 2024b).”
• Kimi-VL Technical Report (2025, neutral): “5-VL (Bai et al. 2025) and Gemma-3 (Gemma Team et al. 2025), continue to rely on dense architectures and do not support long-CoT reasoning.”
• VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model (2025, neutral): “To-day, some of the most widely adopted open-source VLM series include LLaVA[25, 30, 31], QwenVL[6, 50], and In-ternVL [10, 11].”
• Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models (2025, positive): “5-VL-7B-Instruct [31] as the base model and train it via supervised fine-tuning (SFT) for 2 epochs on 32 NVIDIA H800 80G GPUs for ∼ 10 hours using Llama-Factory framework [51].”
• Qwen2.5-Omni Technical Report (2025, neutral): “5-VL (Bai et al., 2025), which is based on the Vision Transformer (ViT) model with approximately 675 million parameters, enabling it to effectively handle both image and video inputs.”
• GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning (2025, neutral): “Over the past decade, as the intelligence level of models has advanced dramatically [36; 47; 20; 4], the complexity of corresponding multimodal intelligence tasks has increased accordingly.”
• Qwen-Image Technical Report (2025, positive): “…extend our base model to explore multi-modal image generation tasks that incorporate both text and image inputs, including instruction-based image editing (Wang et al., 2025), novel view synthesis (Wang et al., 2024b), and computer vision tasks such as depth estimation (Bochkovskii et al., 2024).”
• R1-Onevision: Advancing Generalized Multimodal Reasoning Through Cross-Modal Formalization (2025, neutral): “5-VL [2], despite its strong multimodal capabilities, lacks deep reasoning ability and ultimately fails to solve the problem.”
• Emerging Properties in Unified Multimodal Pretraining (2025, positive): “5-VL-7B [4] and insert it directly before the image as a conceptual scaffold.”
• VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning (2025, neutral): “5-VL-72B [Bai et al., 2025], QvQ-72B [Wang et al., 2024b], InternVL-2.”
• Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (2025, neutral): “In parallel, reinforcement learning has also gained attention in the multimodal domain, driving advancements in multimodal reasoning (Chen et al., 2025; Shen et al., 2025; Zheng et al., 2025).”
• InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models (2025, neutral): “Yet, the majority of leading MLLMs—both open-source and proprietary—are adapted from text-only large language models through sophisticated multi-stage pipelines [21, 19, 18, 5, 121, 7].”
• Step1X-Edit: A Practical Framework for General Image Editing (2025, positive): “5-VL [3 ] and the Recognize-Anything Model [69] to identify target objects or keywords, followed by Flux-Fill [6] for content-aware inpainting.”
• Qwen3 Technical Report (2025, positive): “5-VL model (Bai et al., 2025) to perform text recognition on a large volume of PDF-like documents.”
• Group-in-Group Policy Optimization for LLM Agent Training (2025, neutral): “5-VL-3B-Instruct [ 66].”