Tether AI Research on Wednesday announced the release of its open-source translation models designed to run on smartphones, laptops, and other devices without an internet connection.
The models cover 19 African languages and nine European languages. They process translations locally, keeping users’ data on their devices instead of sending it to third-party cloud servers, according to a statement.
The African languages models include TranslatePsy-AfriSLM, which covers Hausa, Amharic, Yoruba, Swahili and 15 other languages, and TranslatePsy-AfriNano, which supports eight African languages. Tether said the 19 languages supported by AfriSLM collectively represent roughly half of Africa’s population.
Notably, AfriSLM’s smallest model has 800 million parameters and outperformed Qwen3.5-122B-A10B, TranslateGemma-27B, and NLLB-3.3B in multiple translation benchmarks, the company said.
Tether said the models can be used for education, healthcare, agriculture and humanitarian work in areas where internet access is limited. AfriSLM can also be paired with QVAC MedPsy, its smaller model for medical and healthcare applications, for health education in local languages.
For Europe, TranslatePsy-EuroNano supports nine languages and 90 translation directions, using English as a pivot. The highest-quality EuroNano model retained 98.4% of Meta’s NLLB-200 translation quality when translating into English, Tether said.
The TranslatePsy models are available through Tether’s collection on Hugging Face, including AfriSLM versions with 0.8 billion, 2 billion, and 4 billion parameters, as well as the EuroNano and AfriNano models.
The research behind TranslatePsy-AfriSLM has also been accepted for presentation at the Empirical Methods in Natural Language Processing 2026 conference, according to the statement.