Tether AI Research has announced the release of a new family of open-source AI translation models designed to run directly on everyday devices without requiring internet connectivity. The launch includes QVAC TranslatePsy-AfriSLM and QVAC TranslatePsy-AfriNano for African languages, as well as QVAC TranslatePsy-EuroNano for European languages, marking a significant expansion of Tether’s broader QVAC artificial intelligence initiative.
The announcement is notable not only for the technology itself, but for its stated objective: making AI-powered translation accessible to populations that have historically been underserved by digital infrastructure. By enabling translation directly on smartphones, laptops, and edge devices, the models are designed to operate privately and offline, keeping user data on local devices rather than relying on third-party cloud services.
At the center of the release is TranslatePsy-AfriSLM, which supports 19 African languages, including Hausa, Amharic, Yoruba, Swahili, Igbo, Somali, Zulu, Xhosa, Kinyarwanda, and Afrikaans. According to Tether AI Research, these languages collectively represent roughly half of Africa’s population, extending the reach of AI-powered communication and content access across West, East, Central, and Southern Africa.
Tether highlighted the efficiency of the new models as a key achievement. Despite containing only 800 million parameters, the smallest TranslatePsy-AfriSLM model reportedly outperformed much larger translation systems across recognised translation benchmarks, including FLORES-200, BOUQuET and SMOL. The company attributes this performance to a new quality-estimation filtering approach that removes up to 96% of low-quality open-source training data, allowing the models to achieve stronger results with significantly less computational overhead.
The announcement positions the technology as a potential enabler of wider access to educational resources, scientific content and AI-driven learning tools in local languages. Tether argues that language remains one of the most significant barriers to AI adoption and that locally deployable translation models can help bridge this divide, particularly in regions where reliable connectivity remains limited.
Alongside the African-focused models, Tether also introduced TranslatePsy-EuroNano, a multilingual translation system covering nine European languages. Using English as a pivot language, the solution supports 90 translation directions while maintaining a highly compact footprint. The company stated that its smallest deployment requires only 36MB of storage, significantly reducing device requirements compared with existing offline translation solutions.
Commenting on the launch, Tether CEO Paolo Ardoino described the initiative as part of a broader effort to extend the benefits of artificial intelligence beyond traditionally connected and well-resourced markets. He emphasized the importance of ensuring that language does not become a barrier to accessing education, information and AI-powered tools.
The models have been released as open source through Hugging Face, with multiple deployment sizes available to accommodate different hardware requirements. Tether also noted that the research underpinning TranslatePsy-AfriSLM has been accepted for presentation at the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP), underscoring the company’s ambition to establish a meaningful presence in applied AI research.
For executives tracking developments in artificial intelligence, the announcement signals Tether’s growing commitment to localized, decentralized AI infrastructure. Rather than focusing solely on larger cloud-based systems, the company is positioning itself around compact, efficient models capable of extending advanced AI capabilities to communities and devices beyond the reach of traditional digital ecosystems.
Read the full announcement here>>> https://tether.io/news/tether-releases-open-source-ai-translation-models-for-african-and-european-languages/