Techinika
Home
EventsOpportunitiesForumContact Us
HomeEventsOpportunitiesForumContact UsAll Posts

Blog Categories

View All
All
Techinika

A global voice for technology, startups, and innovation. Covering the latest trends, news, and know-how for a global audience from Rwanda.

Quick Links

CompanyPricingAbout Techinika NewsEcosystem CompaniesDid You Know FactsTech InsightsContact Us

Categories

Techinika SearchAll

Stay Updated

Subscribe to our newsletter to get the latest tech news delivered to your inbox.

© 2020 - 2026 Techinika. All rights reserved.

  • Privacy
  • Terms of Use
HomeEventsOpportunitiesForumContact Us
Share this article

  Stay Updated!

Follow our official WhatsApp Channel for instant alerts on new grants, tenders, and job postings.

Follow WhatsApp Channel

Vambo AI Introduces MORENA, A New Large Language Model for African Languages

By Cishahayo Songa Achille•2 minutes ago•5 min read
Vambo AI Introduces MORENA, A New Large Language Model for African Languages

AI Quick Summary

Vambo AI has announced the release of MORENA, a new multilingual AI model developed specifically for African languages. Unveiled on September 18, 2026, MORENA is a 1.5 billion-parameter language model that was trained from scratch, rather than being a fine-tuned version of an existing model.

The initiative aims to provide a robust foundation for building AI applications that are deeply integrated with the linguistic diversity of the African continent.

What is MORENA?

MORENA is a 1.5 billion-parameter language model developed by Vambo AI. It was trained on 12 African languages: ChiShona, Kiswahili, Hausa, Yorùbá, Igbo, isiZulu, isiXhosa, Kinyarwanda, Setswana, Afrikaans, isiNdebele, and Nigerian Pidgin.

A key distinction of MORENA is its development approach. The model was trained from the ground up, featuring its own tokenizer and a data mixture specifically designed for the African languages it supports.

Performance and Accessibility

According to Vambo AI, MORENA demonstrates competitive performance against models from companies such as Google, Meta, Alibaba, and HuggingFace in African language modeling and translation tasks. It is also reportedly up to eight times smaller than some of these comparative models.

On held-out African language text, MORENA achieved a bits per byte (bpb) score of 1.408 for its base version and 1.441 for its instruct version, which Vambo AI states is the lowest among 26 measured models. For translation, the model achieved 45.8 chrF++ when translating English into five African languages.

Read Also Digital Umuganda Releases First Open-Source African Language AI Datasets for Afrivoice V2

Key Fact: MORENA is the largest African language model trained from scratch, featuring 1.5 billion parameters and supporting 12 African languages. Its weights are open and free for commercial use, fostering further research and development.

Development and Support

The training of MORENA involved over 22,000 A100 GPU hours, representing approximately $40,000 worth of GPU compute. The model was trained on the CINECA Leonardo system.

The project received support from the United Nations Development Programme (UNDP), AIHub4SD, and CINECA, which provided the necessary GPU resources and assistance.

Availability for Researchers

Vambo AI has made the MORENA model weights open and free for commercial use. This allows researchers and developers to build upon the model without licensing restrictions. The weights are available on the Hugging Face platform.

In addition to the 1.5 billion-parameter version, Vambo AI has also released 0.5 billion and 0.2 billion-parameter versions of MORENA, including instruct, base, mini, and nano models, designed for various applications like chat, translation, and tool calling.

If you enjoyed this article, follow us on WhatsApp for daily tech updates. If you have an idea, need to be featured or need to partner, reach out to us at editorial@techinika.com or use our contact page.

Community Discussion

Don't let the story end here.

Share your thoughts, ask questions, and connect with the community.

Join the Conversation
About the Author
Cishahayo Songa Achille

Cishahayo Songa Achille

Chief Editor

Cishahayo Songa Achille is a prominent Rwandan software engineer, tech entrepreneur, and digital content creator.

View all articles by Cishahayo Songa Achille →

Up Next

Digital Umuganda Releases First Open-Source African Language AI Datasets for Afrivoice V2

By Cishahayo Songa Achille • 5 min read

Stay Updated

Get the latest tech insights delivered to your inbox. No spam, just quality content.

We respect your privacy. You can unsubscribe at any time.

Similar Articles

Digital Umuganda Releases First Open-Source African Language AI Datasets for Afrivoice V2

Digital Umuganda Releases First Open-Source African Language AI Datasets for Afrivoice V2

Cishahayo Songa Achille • 15 days ago • 5 min read
AI Is Speeding Up Open Source Security, But Developers Still Have the Final Say

AI Is Speeding Up Open Source Security, But Developers Still Have the Final Say

ISHIMWE Jean Claude • about 1 month ago • 3 min read
Rwanda and Google Explore New AI Partnership as Country Pushes to Build Its AI Economy

Rwanda and Google Explore New AI Partnership as Country Pushes to Build Its AI Economy

ISHIMWE Jean Claude • about 1 month ago • 4 min read