Vambo AI and Tether Release African-Language AI Models, With MORENA Beating Larger Rivals

MORENA covers Nigerian Pidgin, Igbo, Yoruba, Hausa, Kiswahili, ChiShona, isiZulu, isiXhosa, Kinyarwanda, Setswana, Afrikaans and isiNdebele. On a key benchmark, it scored 1.408 bits per byte, the best of 26 models tested, according to Vambo AI. Its closest rival, Lugha-Llama-8B, an Africa-adapted Llama 3.1 variant with more than five times as many parameters, scored 1.423 bpb and lost to MORENA in eight of the 12 languages. A 12B Gemma system consumes roughly 11 times the compute of MORENA while producing a weaker score on the same text, the company says.

Vambo AI says MORENA was built from scratch for African languages rather than continuing training on an existing Llama variant. The developers fixed the tokenizer, data mixture and language list before training began, after comparing vocabulary sizes for cost and efficiency. MORENA's vocabulary encodes African text with 1.39 times fewer tokens than Gemma 3 and 1.53 times fewer than Llama 3.2 on identical passages. African text still costs 0.249 tokens per byte against 0.234 for English, roughly 6% more, a difference the team says it cannot fully explain.

The chat-tuned instruct version of MORENA scores 1.441 bpb, trailing Lugha-Llama-8B overall but leading it in five shared languages while using about 20% of its parameters. After seeing three sample translations, the instruct model renders English into five African languages at 45.8 chrF++, statistically level with one dedicated translation system. It trails a larger translation model by about 1.4 points, while general models of similar size typically land in a range from 9 to 14.

MORENA used 251.7 billion tokens during pretraining and another 63 billion during mid-training. Development required more than 22,000 A100 GPU hours, with computing resources valued at roughly $40,000, according to the developers. Vambo AI says the project was supported by UNDP, AIHub4SD and CINECA with computing resources and other assistance. Smaller releases include 0.5B and 0.2B parameter versions. The 0.5B variants reportedly exceed every external system tested across 11 of the 12 languages. The 0.2B nano version is intended for speech-recognition rescoring, keyboard applications and text normalisation, and costs 58 times less than Lugha-Llama-8B for each byte processed while generating 104 tokens per second on one A100 GPU. A CPU-compatible build runs offline on a laptop.

Tether AI Research's release includes QVAC TranslatePsy-AfriSLM for 19 African languages and QVAC TranslatePsy-EuroNano for nine European languages. Each model processes translations directly on the device, keeping personal data local rather than sending it to remote cloud servers. The African release covers Hausa, Amharic, Yoruba, Lingala, Swahili, Igbo, Zulu, Somali, Oromo, Malagasy, Kinyarwanda, Xhosa, Afrikaans, Wolof, Luganda, Nyanja, Shona, Tswana and Southern Sotho. Tether says those languages collectively account for about 50% of the continent's population, a figure representing language reach rather than individual users.

AfriSLM contains 800 million parameters, and Tether says the model surpassed Qwen3.5-122B-A10B, TranslateGemma-27B and NLLB-3.3B. The comparisons were measured using FLORES-200, BOUQuET and SMOL, although benchmark performance does not establish identical results across every practical translation situation. Tether attributes the results partly to a data-screening system that discarded as much as 96% of material judged unsuitable. The company also released 0.8B, 2B and 4B parameter configurations for different hardware requirements.

EuroNano takes another approach, using English as an intermediary language while connecting European languages through 90 possible translation routes. It occupies only 36MB, which Tether says is about 94% less storage than a comparable Firefox offline arrangement. The compact footprint means translation can remain available after connectivity disappears, without requiring separate downloads for every language combination, a distinction Tether says matters for field workers, travellers and local applications where storage and network availability can impose constraints.

Tether's African focus also connects with physical infrastructure it has developed across parts of Sub-Saharan Africa. The company has installed solar-powered kiosks that provide phone charging, battery exchanges and access to digital financial services. Tether says those locations could eventually provide another route for distributing educational material, agricultural information and other locally translated content without depending on continuous connectivity.

“Four billion people were left behind by the traditional financial system, and the most powerful technology of our age has repeated that failure,” said Paolo Ardoino, CEO of Tether. “Language should not determine who can benefit from artificial intelligence. Open translation models like these are a step toward a future where education and AI tools reach hundreds of millions of people who have neither reliable connectivity nor access to expensive systems.” Ardoino added that the same technology could operate alongside QVAC MedPsy, Tether's smaller model intended for healthcare-related applications on local devices. “A mother could get real medical information she understands, instead of guessing. A child could learn in their own language. That is the future we are building through QVAC,” he said.

AfriSLM is currently available through Hugging Face, and the associated research has been accepted for presentation at EMNLP 2026.

Editor's Summary

Vambo AI's MORENA and Tether's offline translation models expand AI support for African languages while reducing computing and connectivity requirements. Vambo AI says MORENA outperforms larger models on several benchmarks, while Tether says its on-device systems keep data local and can operate without network access. Both releases include smaller configurations aimed at devices with limited resources.