Cohere, an Enterprise AI company, has recently unveiled a new family of multilingual models called Tiny Aya at the India AI Summit. These models are open-weight, allowing anyone to access and modify their underlying code. They support over 70 languages and can be run on everyday devices like laptops without the need for an internet connection.
The research arm of Cohere, Cohere Labs, has specifically designed the Tiny Aya models to cater to South Asian languages such as Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi. The base model boasts an impressive 3.35 billion parameters, showcasing its size and complexity. Additionally, Cohere has introduced specialized versions like TinyAya-Global for broad language support, TinyAya-Earth for African languages, TinyAya-Fire for South Asian languages, and TinyAya-Water for Asia Pacific, West Asia, and Europe.
According to the company, this approach allows each model to have a stronger linguistic foundation and cultural nuance, making the systems more natural and reliable for the communities they serve. Despite the specific regional focus, all Tiny Aya models maintain broad multilingual coverage, providing a flexible starting point for further customization and research.
One of the key highlights of these models is their ability to run directly on devices, enabling offline translation capabilities. This feature makes them ideal for developers and researchers creating apps for native language-speaking audiences. Cohere has optimized the underlying software for on-device usage, requiring less computing power compared to similar models in the market.
In countries like India with diverse linguistic populations, the offline-friendly nature of these models opens up a wide range of applications and use cases without the constant need for internet connectivity. Developers can access the models on platforms like HuggingFace, Kaggle, and Ollama for local deployment. Cohere is also releasing training and evaluation datasets on HuggingFace and plans to publish a technical report detailing its training methodology.
Aidan Gomez, the CEO of Cohere, has hinted at the company’s plans to go public in the near future. Reports from CNBC indicate that Cohere had a successful year in 2025, ending with $240 million in annual recurring revenue and maintaining a 50% growth quarter-over-quarter throughout the year. With the launch of the Tiny Aya models, Cohere continues to make significant strides in the AI industry, offering innovative solutions for multilingual communication and cultural understanding.
