Our Approach

Qwen began as part of Alibaba’s long-term investment in cloud and AI infrastructure. The focus has been simple: build strong foundations first, then turn advanced models into tools that people can actually use.

Alibaba published its first open weight models, Qwen and Qwen Chat, in August 2023. The first-generation Qwen models’ code, weights, and documentation were made freely accessible to academics, researchers, and commercial institutions worldwide on Hugging Face and Alibaba’s open-source community platform, ModelScope.

Over time, Qwen has evolved into a family of models that support everyday tasks, from writing and coding to understanding images, audio, and data. The emphasis has been on usefulness, cost-efficiency, and integration into existing systems, so that organizations and developers can adopt AI in a steady and manageable way, while maintaining oversight and following responsible practices.

Rather than treating AI as a standalone product, Qwen has been developed alongside cloud infrastructure and applications. This full-stack approach helps ensure performance, scalability, operational reliability, and the safe application of AI in real-world environments.

From the start, Qwen has taken an open approach. Many of Qwen's models today are open weight, allowing researchers, developers, and companies around the world to build on them, adapt them to local languages, and tailor them to specific needs, while retaining guidance on safe and responsible use.

Today, Qwen models are used across a wide range of countries and sectors. Organizations and developers in different regions have adapted the models for business and public services, research, applications in a variety of sectors. This global community helps strengthen the models over time, encourages responsible practices, and ensures they reflect diverse languages and use cases.

By lowering technical and cost barriers, Qwen aims to make advanced AI more accessible in a way that is practical, accountable, and safe, not limited to a small number of institutions.

The Qwen model series is designed with deep commitment to linguistic inclusivity, supporting 201 languages via the Qwen3.5 model family—not only major global languages but also many regional, minority, and historically underrepresented ones. This includes languages such as Swahili, Khmer, Lao, Nepali, Telugu, Bengali, Urdu, and even many regional dialects.

By embracing this rich tapestry of languages, Qwen ensures that communities often excluded from the AI revolution can access, interact with, and benefit from advanced technology in their own linguistic and cultural contexts. This approach reflects our belief that the progress of AI must be inclusive, equitable, and accessible to all, regardless of geography or language size.

Qwen is designed to be useful, but also to be deployed carefully. Clear usage standards, technical safeguards, and human oversight are part of how the system is developed and managed.

The goal is to support decision-making, not replace it. In higher-risk or regulated settings, responsibility remains with the organizations using the system. Data protection, transparency, and compliance with applicable laws are treated as core requirements, not afterthoughts.

As AI continues to evolve, Qwen’s approach remains focused on three principles:
· Global access
· Practical value
· Responsible deployment