African AI Talent Shrinks Continent’s First Multilingual Language Model by 75%

In a major leap for inclusive and accessible artificial intelligence, Africa’s first multilingual Small Language Model (SLM), InkubaLM, has been reduced in size by an impressive 75%, without compromising performance. This milestone was achieved through the Buzuzu-Mavi Challenge, a global AI competition hosted by Lelapa AI in partnership with Zindi.

The challenge invited machine learning experts to compress InkubaLM, which is designed to support African languages, making it more efficient and easier to deploy in low-resource environments. Out of 490 participants from 61 countries, the top winners, all from Africa—demonstrated the continent’s growing strength in AI innovation.

Yvan Carré from Cameroon took first place, applying a mix of adapter heads, quantisation, and knowledge distillation to dramatically slim down the model. South Africa’s Stefan Strydom secured second place by cutting the model to just 40 million parameters through vocabulary trimming and shared embeddings. The AI_Buzz team, Abdourahamane Ide Salifou, Mubarak Muhammad, and Victor Olufemi from Niger and Nigeria, won third, using blended datasets and model distillation to create an efficient student model.

This development is especially significant for Africa, where only about 33% of the population has regular internet access and 70% rely on entry-level smartphones. Lean AI models like InkubaLM can operate on these devices without constant connectivity, powering practical solutions in sectors such as education, agriculture, translation, and customer service.

“This challenge isn’t simply about technical progress, it reflects our mission to build AI that’s inclusive and grounded in African realities,” said Pelonomi Moiloa, CEO of Lelapa AI. “When African talent is trusted with meaningful challenges, the results are transformative.”

Celina Lee, CEO of Zindi, added: “We’re proud to show the world that more can be done with less. These solutions showcase what African innovators can achieve in the language model space.”

Looking ahead, the most promising models from the challenge will inform future versions of InkubaLM, which remains open-source and available to the global tech community. Lelapa AI and Zindi are inviting continued collaboration to make African AI even smaller, smarter, and more impactful.