Cohere releases Tiny Aya, a family of open-weight multilingual models

Enterprise artificial-intelligence company Cohere, through its research arm Cohere Labs, has launched Tiny Aya, a new family of open-weight AI models—meaning their underlying code is publicly available for anyone to use and modify. The models support over 70 languages and can run on everyday devices, such as laptops, without requiring an internet connection. This offline capability is designed to enable developers to build and run applications locally in linguistically diverse regions.

The base model in the family contains 3.35 billion parameters, which is a measure of the model’s size and complexity. Cohere trained the models on a cluster of 64 H100 GPUs manufactured by Nvidia, utilizing relatively modest computing resources. Alongside the base model, Cohere released TinyAya-Global, a version fine-tuned to better follow user commands for applications requiring broad language support. The family also includes three distinct regional variants:

• TinyAya-Earth: Designed for African languages.
• TinyAya-Fire: Designed for South Asian languages, supporting languages in India such as Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi.
• TinyAya-Water: Designed for languages across the Asia Pacific, West Asia, and Europe.

Cohere explained the strategy behind the regional variants: “This approach allows each model to develop stronger linguistic grounding and cultural nuance, creating systems that feel more natural and reliable for the communities they are meant to serve. At the same time, all Tiny Aya models retain broad multilingual coverage, making them flexible starting points for further adaptation and research.” Developers can download the models for local deployment on HuggingFace, Kaggle, and Ollama. The company is also releasing training and evaluation datasets on HuggingFace.

The release comes during a period of financial growth for the startup. According to reporting by CNBC, Cohere ended 2025 with $240 million in annual recurring revenue, maintaining a 50% growth rate quarter-over-quarter throughout the year. Looking ahead, Cohere CEO Aidan Gomez has stated that the company plans to go public soon.

Why it matters

Cohere’s release of the open-weight Tiny Aya model family enables on-device, offline AI capabilities for over 70 languages, targeting developers and researchers in linguistically diverse regions.