Why Open Weight?

Open access as a public benefit

Open access as a public benefit

We believe advanced AI should be widely accessible, not limited to a small number of large companies. Sharing model weights allows developers to build on existing progress rather than starting from zero. This lowers costs, speeds up innovation, and supports fairer competition – especially for SMEs and startups.

Training a large language model with more than 100 billion parameters is costly, considering data, computing power, and talent. Most small organizations and individual developers cannot afford this investment alone. By absorbing these costs and releasing model weights openly, we make advanced AI available at a fraction of the original expense. This creates broad economic value and reduces barriers to entry across the industry.

To support this, many of our models are released under the Apache License 2.0. This license is approved by the Open Source Initiative. It allows anyone to use, modify, and distribute the software for research or commercial purposes without fees. Users must retain copyright and attribution notices, indicate any changes, and include the license. They are not required to publish their own code built on top of the models, and contributors cannot later bring patent claims against compliant users.

Openness builds trust

Openness builds trust

Public access to model parameters, documentation, and evaluation tools enables broad inspection and independent verification. Researchers, companies, and civil society can examine how systems work and assess their performance across different scenarios. This visibility helps identify vulnerabilities early and improves safety, reliability, and alignment with human values.

By combining open collaboration with clear safeguards, we aim to expand access, support innovation, and ensure advanced AI develops in a secure, transparent, and inclusive way.

Practical flexibility across industries

Practical flexibility across industries

Open models can be adapted to different technical and business needs. Developers can fine-tune them for specific tasks, adjust architectures, or integrate them into sector-specific systems.

Different use cases require different model sizes. Some applications need lightweight models that run efficiently on devices with limited storage and computing power. Others require large, high-performance systems. By offering a range of model sizes — from compact versions to sophisticated Mixture-of-Experts systems — we support diverse industries including education, healthcare, manufacturing, and scientific research.

Protection, privacy, and control

Protection, privacy, and control

Open weight models give organisations greater operational control. They can fine-tune models using sensitive data within secure internal environments, rather than transmitting that data to external APIs. This supports privacy protection, customisation, and regulatory compliance in sensitive sectors.

Balancing openness and safety

Balancing openness and safety

Our most capable models are made available only through our proprietary AI tools or API via Alibaba Cloud Model Studio. This is done with the aim to strike a balance between model safety and the technological benefits stemming from public access to these models. Providing access through a managed API allows real-time monitoring, rule enforcement, rapid response to emerging threats, and continuous safety updates without affecting legitimate users.

Maintaining proprietary control over our most advanced models supports this balanced approach. It enables us to make many other high-quality models freely available on platforms such as Hugging Face and GitHub, where they have become among the most downloaded open-source models in the world.