![[2510.14979] From Pixels to Words -- Towards Native Vision-Language Primitives at Scale](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2510.14979] From Pixels to Words -- Towards Native Vision-Language Primitives at Scale
Summary
The paper discusses the development of native Vision-Language Models (VLMs) that integrate vision and language capabilities more effectively than traditional modular approaches, presenting a new model called NEO.
Why It Matters
This research addresses critical challenges in the field of AI, particularly in enhancing the integration of visual and textual data. By proposing a novel framework for native VLMs, it aims to democratize access to advanced AI technologies and accelerate innovation in vision-language applications.
Key Takeaways
- Native VLMs can outperform modular models by integrating vision and language more cohesively.
- The NEO model is designed to align pixel and word representations within a shared semantic space.
- The research emphasizes the importance of accessibility in advancing native VLM technologies.
- NEO utilizes a large dataset of 390M image-text pairs to enhance visual perception.
- The paper provides reusable components for building scalable native VLMs.
Computer Science > Computer Vision and Pattern Recognition arXiv:2510.14979 (cs) [Submitted on 16 Oct 2025 (v1), last revised 21 Feb 2026 (this version, v2)] Title:From Pixels to Words -- Towards Native Vision-Language Primitives at Scale Authors:Haiwen Diao, Mingxuan Li, Silei Wu, Linjun Dai, Xiaohua Wang, Hanming Deng, Lewei Lu, Dahua Lin, Ziwei Liu View a PDF of the paper titled From Pixels to Words -- Towards Native Vision-Language Primitives at Scale, by Haiwen Diao and 8 other authors View PDF HTML (experimental) Abstract:The edifice of native Vision-Language Models (VLMs) has emerged as a rising contender to typical modular VLMs, shaped by evolving model architectures and training paradigms. Yet, two lingering clouds cast shadows over its widespread exploration and promotion: (-) What fundamental constraints set native VLMs apart from modular ones, and to what extent can these barriers be overcome? (-) How to make research in native VLMs more accessible and democratized, thereby accelerating progress in the field. In this paper, we clarify these challenges and outline guiding principles for constructing native VLMs. Specifically, one native VLM primitive should: (i) effectively align pixel and word representations within a shared semantic space; (ii) seamlessly integrate the strengths of formerly separate vision and language modules; (iii) inherently embody various cross-modal properties that support unified vision-language encoding, aligning, and reasoning. Hence, ...
