Cohere Labs has released Tiny Aya, a multilingual AI model designed to address the limitations of existing language models by supporting over 70 languages and running efficiently on local devices. The model is open-weight, allowing developers to create practical AI solutions without requiring powerful hardware. This innovation aims to expand access to underrepresented languages and improve the usability of AI in real-world scenarios.
Projects developed using Tiny Aya have yielded insights across four key themes: education and learning, building safer multilingual AI, accessibility and local deployment, and language understanding and processing. These initiatives highlight how accessible models, combined with strong mentorship, can drive meaningful AI innovation. For instance, teams have created tools to make learning more accessible and developed child-safe AI voice assistants that operate entirely offline.
The research behind Tiny Aya was recently accepted to COLM 2026, underscoring the model's impact as both a practical foundation for multilingual AI and a contribution to the research community. The model's ability to rival larger models in tasks like word-sense disambiguation and its sensitivity to programming language syntax further demonstrate its capabilities.
Source: cohere
