Tether Releases Open-Source AI Translation Model, Supporting 19 African Languages with Offline Capability

Odaily News - Tether AI Research, the AI research team under Tether, announced the release of open-source AI translation models QVAC TranslatePsy-AfriSLM, QVAC TranslatePsy-AfriNano, and QVAC TranslatePsy-EuroNano, supporting 19 African languages, 8 African languages, and 9 European languages respectively. These models can run offline locally on devices such as smartphones and laptops.

The TranslatePsy models can handle translations without requiring an internet connection, ensuring user data remains on the device. With only 800 million parameters, the TranslatePsy-AfriSLM outperforms Qwen3.5-122B-A10B, TranslateGemma-27B, and NLLB-3.3B on the FLORES-200, BOUQuET, and SMOL benchmarks.

TranslatePsy-EuroNano supports 90 translation directions among 9 European languages, with the smallest deployment version requiring only 36MB of storage—approximately 94% less than a comparable Firefox offline translation setup. TranslatePsy-AfriSLM is now available for download on Hugging Face, and the related research has been accepted by EMNLP 2026. (Tether News)