Cohere Labs, an American-Canada-based international technology company focused on AI, has launched “Tiny Aya,” a new family of open-weight multilingual models designed to operate on standard hardware without an internet connection.
According to their statement, the 3.35-billion-parameter models support over 70 languages and are engineered to provide high-performance AI capabilities in resource-constrained environments. By utilizing a “right-sized” architecture, Tiny Aya enables sophisticated tasks such as translation, summarization, and conversational AI to run locally on devices like laptops, bridging the digital divide for underrepresented language communities.
The release features a base global model alongside specialized regional variants. This regional approach is intended to strengthen linguistic grounding and cultural nuance, ensuring AI interactions feel more natural to the communities they serve.
Developed by Cohere Labs using a cluster of 64 Nvidia H100 GPUs, the models were created with a focus on resource efficiency. Tiny Aya is currently available for download on Hugging Face, Kaggle, and Ollama for research and local implementation.
