South Africa's Vambo AI Releases Open-Weight LLM Trained on 12 African Languages
South African company Vambo AI has released an open-weight large language model trained from scratch across 12 African languages, a rare instance of an African-built foundation model entering the market. The release is being noted as a significant milestone for African-language AI capability.
MORENA: proving African-language AI doesn't need a billion-dollar budget
The most striking claim in Vambo AI's release isn't the 12 languages — it's the efficiency argument underneath them.
MORENA's 1.5 billion parameters reportedly beat a Google model eight times its size on African text. The reason: it tokenises African languages more efficiently, meaning it costs less to run and less to adapt. A Nigerian startup fine-tuning it for Hausa spends a fraction of what it would building on a Western foundation model.
That changes the economics of who can build here. The barrier to African-language AI has never been imagination — it's been the cost of entry. A smaller, cheaper, open model lowers that floor in a way the telecoms-led approach simply can't: no revenue-sharing agreement, no operator gatekeeping, just a model anyone can take and build on.
1 source
- SA’s Vambo AI releases language model trained for 12 African languages · disruptafrica.com · T1