Cohere Unveils Open-Source Multilingual AI Model Family

Enterprise AI firm Cohere has unveiled a new family of multilingual models, named Tiny Aya, during the ongoing India AI Summit. These open-weight models—meaning their core code is publicly accessible for use and modification—support over 70 languages and can operate on everyday devices like laptops without an internet connection.

Developed by Cohere's research division, Cohere Labs, the model includes support for South Asian languages such as Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi.

The base model features 3.35 billion parameters, a key indicator of its scale and sophistication. Cohere also introduced TinyAya-Global, a version fine-tuned for improved instruction-following, ideal for applications needing wide linguistic coverage. The family is completed with regional variants: TinyAya-Earth for African languages; TinyAya-Fire for South Asian languages; and TinyAya-Water for Asia Pacific, West Asia, and Europe.

"This approach enables each model to develop deeper linguistic understanding and cultural nuance, creating systems that feel more natural and dependable for their target communities. Simultaneously, all Tiny Aya models maintain broad multilingual capabilities, serving as flexible foundations for further adaptation and research," the company stated.

Cohere highlighted that these models, trained on a single cluster of 64 NVIDIA H100 GPUs with relatively modest computational resources, are perfect for researchers and developers building applications for native language speakers. The models run directly on devices, enabling offline translation features. The company built its underlying software specifically for on-device use, requiring less computing power than most comparable models.

In linguistically diverse nations like India, this offline capability unlocks a wide range of applications and use cases without relying on constant internet connectivity.

The models are available on Hugging Face, the popular platform for sharing and testing AI models, and the Cohere Platform. Developers can download them from Hugging Face, Kaggle, and Ollama for local deployment. Cohere is also releasing training and evaluation datasets on Hugging Face and plans to publish a technical report detailing its training methodology.

TechCrunch Founder Summit 2026: Tickets Now Available

On June 23 in Boston, over 1,100 founders will gather at the TechCrunch Founder Summit 2026 for a full day dedicated to growth, execution, and real-world scaling. Gain insights from founders and investors who have defined the industry. Connect with peers facing similar growth challenges. Leave with actionable tactics you can implement right away.

Save up to $300 on your pass or save up to 30% with group tickets for teams of four or more.

TechCrunch Founder Summit: Tickets Now Available

On June 23 in Boston, over 1,100 founders will gather at the TechCrunch Founder Summit 2026 for a full day dedicated to growth, execution, and real-world scaling. Gain insights from founders and investors who have defined the industry. Connect with peers facing similar growth challenges. Leave with actionable tactics you can implement right away.

Save up to $300 on your pass or save up to 30% with group tickets for teams of four or more.

Cohere's CEO, Aidan Gomez, stated last year that the company intends to go public "soon." According to CNBC, the company finished 2025 strongly, reporting $240 million in annual recurring revenue with 50% quarter-over-quarter growth throughout the year.