South African startup Vambo AI has released MORENA, an artificial intelligence language model built from scratch to support 12 African languages, alongside English and French. The release gives developers and researchers access to a foundation they can adapt for African-language applications.
African Languages at the Centre
MORENA’s language coverage includes ChiShona, Kiswahili, Hausa, Yorùbá, Igbo, isiZulu, isiXhosa, Kinyarwanda, Setswana, Afrikaans, isiNdebele and Nigerian Pidgin.
The 1.5-billion-parameter model was developed with its own training-data mixture and tokenizer—the system that breaks text into units an AI model can process. Vambo’s research places these African languages at the centre of the model’s design.
For South African developers, its coverage creates opportunities to explore digital services in several local languages, including customer-support tools, educational applications and translation services. These are potential uses that would require adaptation and testing for their intended audiences.
Available for Developers to Build On
MORENA is an open-weight release, meaning its trained model parameters are available for others to download and adapt. Its Apache 2.0 licence permits commercial use, fine-tuning and redistribution under the licence’s terms, although the training corpus has not been released with the weights.
Vambo distinguishes between the base model, intended as a starting point for further training, and an instruction-tuned version designed for tasks such as chat, translation and tool calling. Smaller models are also listed in the MORENA family, offering additional options for developers working with different computing requirements.
Promising Results, With Specific Limits
Vambo reports that MORENA’s base model achieved the strongest result among 26 models measured on its African-language text-modelling evaluation. Its published research also reports competitive English-to-African-language translation results.
These are developer-reported findings on specific tests, rather than evidence that MORENA outperforms larger models across every task.
The model documentation acknowledges trade-offs, including limited general factual knowledge. It also makes clear that the base version is not a ready-to-use chatbot, reinforcing the importance of further training and evaluation before deployment.
By making MORENA available for adaptation, Vambo is giving African developers a foundation for building tools around the languages their users speak—and an opportunity to test how locally focused AI can serve those users.
