Cohere Releases Tiny Aya Open Models for Offline Multilingual AI
Cohere unveils Tiny Aya, a family of open-weight multilingual AI models that run locally on laptops without internet access. The models support over 70 languages and target regional linguistic communities across South Asia, Africa, and Europe.
Enterprise AI company Cohere introduces a new family of open-weight multilingual models called Tiny Aya during the India AI Summit. These models support over 70 languages and run directly on everyday devices like laptops without requiring an internet connection. With a base size of 3.35 billion parameters, the models operate efficiently on relatively modest computing resources, making them highly accessible for local deployment.
The Tiny Aya family includes specialized regional variants to provide stronger cultural nuance and linguistic grounding. TinyAya-Fire focuses on South Asian languages such as Hindi, Bengali, and Tamil, while TinyAya-Earth targets African languages and TinyAya-Water covers the Asia Pacific, West Asia, and Europe. Additionally, Cohere offers TinyAya-Global, a version fine-tuned specifically to follow user commands for applications needing broad language support.
Developers can download these offline-friendly models from popular platforms like HuggingFace, Kaggle, and Ollama. This local capability unlocks diverse applications in linguistically diverse regions like India by eliminating the need for constant internet connectivity. Cohere also releases the underlying training and evaluation datasets to support further adaptation and research by the global AI community.