Cohere Launches New Family of Open Multilingual AI Models
Enterprise AI company Cohere has introduced a new family of multilingual models, named Tiny Aya, at the India AI Summit. These models are open-weight, meaning their code is publicly accessible for users to modify and utilize. They support over 70 languages and are designed to run on common devices like laptops without needing an internet connection.
Developed by Cohere Labs, the research division of Cohere, the Tiny Aya models offer support for numerous South Asian languages, including Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi.
The foundational model features 3.35 billion parameters, indicating its size and complexity. Cohere has also released TinyAya-Global, a variant optimized for following user commands, suitable for applications requiring extensive language coverage. The family includes specialized regional versions: TinyAya-Earth for African languages, TinyAya-Fire for South Asian languages, and TinyAya-Water for regions in Asia Pacific, West Asia, and Europe.
"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," stated the company.
Cohere highlighted that these models were trained using a relatively moderate amount of computing resources on a single cluster of 64 H100 GPUs. They are particularly well-suited for researchers and developers creating applications for native language speakers. The ability of these models to run directly on devices enables offline translation capabilities. Cohere emphasized that the underlying software was engineered for on-device usage, requiring less computational power than many comparable models.
In linguistically rich nations like India, this offline functionality can unlock a wide array of applications and use cases, removing the dependency on constant internet access.
The Tiny Aya models are accessible on HuggingFace, a popular platform for AI model sharing, as well as the Cohere Platform. Developers can download them from HuggingFace, Kaggle, and Ollama for local implementation. Cohere is also releasing the training and evaluation datasets on HuggingFace and plans to publish a technical report detailing their training methods.
The company's CEO, Aidan Gomez, previously indicated plans for an upcoming public offering. Cohere reportedly concluded the past year with strong financial performance, achieving $240 million in annual recurring revenue and experiencing consistent 50% quarter-over-quarter growth throughout the year.
Stay Tuned to Devignitor Insights for More Updates
