Cohere Launches Open-Weight Multilingual AI Models – Tiny Aya

Enterprise AI company Cohere has unveiled a new family of open-weight multilingual models, dubbed Tiny Aya, at the India AI Summit. These models support over 70 languages and are designed to operate on devices like laptops without an internet connection, marking a significant step towards accessible AI for diverse linguistic communities.

Expanding Access to Multilingual AI

The Tiny Aya models, developed by Cohere Labs, aim to bridge the gap in AI accessibility for languages beyond English. The base model boasts 3.35 billion parameters, offering substantial complexity while remaining efficient. Support is included for a wide range of South Asian languages, including Bengali, Hindi, Punjabi, Urdu, Gujarati, Tamil, Telugu, and Marathi. This focus addresses the need for AI tools tailored to the nuances of these languages.

Model Variants for Regional Specificity

Cohere is releasing several variants of Tiny Aya to cater to specific regional needs:

• TinyAya-Global: Fine-tuned for broad language support and improved adherence to user commands.
• TinyAya-Earth: Optimized for African languages.
• TinyAya-Fire: Specifically designed for South Asian languages.
• TinyAya-Water: Tailored for the Asia Pacific, West Asia, and European regions.

According to Cohere, this modular approach allows each model to develop stronger linguistic grounding and cultural nuance, resulting in more natural and reliable systems for the communities they serve. All models retain broad multilingual coverage, making them adaptable for further research and customization.

Efficient Training and Deployment

These models were trained using relatively modest computing resources – a single cluster of 64 Nvidia H100 GPUs – highlighting Cohere’s commitment to efficient AI development. The models are available for download on HuggingFace, Kaggle, and Ollama for local deployment. Cohere is also releasing training and evaluation datasets on HuggingFace, along with a technical report detailing their training methodology.

Cohere’s Growth and Future Outlook

Cohere CEO Aidan Gomez indicated plans for a public offering “soon.” The company reported $240 million in annual recurring revenue at the end of 2025, with 50% quarter-over-quarter growth, demonstrating strong financial performance. CNBC

Key Takeaways

• Cohere’s Tiny Aya models offer open-weight, multilingual AI capabilities.
• The models support over 70 languages, including key South Asian languages.
• Regional variants are available for optimized performance in specific geographic areas.
• The models are designed for efficient on-device operation, enabling offline functionality.
• Cohere is experiencing significant growth and plans to go public.