Orange releases five Wolof AI models built on Alibaba’s Qwen3

What happened
Orange has released five open-weight AI language models for Wolof, a language spoken by around 16 million people in West Africa. The models were fine-tuned from Alibaba's Qwen3 and are available for free non-commercial use on Hugging Face. Orange announced the launch at its OpenTech event in Paris. The models were trained using public datasets that were automatically translated into Wolof, and their quality was checked by Wolof-speaking annotators.
Key details
- Five models released, ranging from 0.6 billion to 14 billion parameters
- Models are based on Alibaba's Qwen3 and available on Hugging Face for non-commercial use
- Training data consisted of public datasets translated into Wolof, with no Orange corporate data used
- Commercial use may be possible through a separate agreement with Orange
- Wolof-speaking annotators evaluated the training data and model outputs for accuracy and naturalness
Why it matters
Wolof speakers in West Africa now have access to AI language models in their own language for free non-commercial use, which can support local developers, researchers, and organizations. Companies or individuals seeking to use these models commercially will need to negotiate with Orange. This release provides a new resource for digital tools and services in Wolof, potentially improving access to technology for millions of people.
