Vambo AI Launches MORENA Model for 12 African Languages

In Short

• South African company Vambo AI launched MORENA, a new AI language model.
• It supports 12 African languages, plus English and French.
• The model is “open-weight”, meaning developers can download and run it, instead of only calling a hosted API.

What Happened

Vambo AI says its new model, MORENA, was trained from scratch with 1.5 billion parameters. Parameters are the internal knobs an AI model learns during training, more parameters can help, but training data and design choices also matter.

MORENA supports ChiShona, Kiswahili, Hausa, Yoruba, Igbo, isiZulu, isiXhosa, Kinyarwanda, Setswana, Afrikaans, isiNdebele, and Nigerian Pidgin. The company positions the release as a foundation model that teams can adapt.

A key technical claim is improved tokenisation. Tokenisation is how a model breaks text into smaller pieces to process it, like chopping a sentence into chunks. Many models built mainly for English use more tokens for some African languages, which can increase compute costs and slow responses.

Vambo AI says MORENA is more efficient on African languages, which could reduce the cost of fine-tuning and running local-language AI in apps like search, translation, transcription, and chatbots.

The co-founder, Isheanesu Misi, also claimed MORENA outperforms a Google model with roughly eight times more parameters on African text modelling. Vambo AI has not published the benchmark results or named the exact Google model, so the comparison cannot be independently verified yet.

Why It Matters

African startups and research groups often rely on global large language models that do not handle local languages well. That can make building customer support bots, voice tools, or public-service applications harder and more expensive.

An open-weight African language model can help teams build products that work better for local users, and can run in regional infrastructure when needed.

There are still open questions. Vambo AI has not publicly shared MORENA’s licensing terms, training datasets, or training compute costs. The company also advises developers to fine-tune the model for specific use cases rather than deploy it as-is.