Cohere Launches Tiny Aya, an Open Multilingual AI Model for the Real World

While most AI models chase English-first scale, Cohere is going the opposite direction.

On the sidelines of the India AI Summit, Cohere unveiled Tiny Aya, a new family of open-weight multilingual AI models designed to run locally, support underserved languages, and work even without an internet connection.

This isn’t a flashy frontier model. It’s infrastructure AI.

What Cohere just launched

The Tiny Aya family is:

• Open-weight — developers can inspect, modify, and deploy it freely

• Multilingual, supporting 70+ languages

• Lightweight, capable of running on everyday devices like laptops

• Offline-ready, removing reliance on constant cloud access

The base model clocks in at 3.35 billion parameters, striking a balance between capability and efficiency.

Built for real-world language diversity

Developed by Cohere Labs, Tiny Aya places special focus on regions typically ignored by big models.

It supports key South Asian languages including Hindi, Bengali, Tamil, Telugu, Marathi, Punjabi, Gujarati, and Urdu — with additional regional variants designed for broader global coverage.

The lineup includes:

• TinyAya-Global — fine-tuned for stronger instruction-following

• TinyAya-Fire — optimized for South Asian languages

• TinyAya-Earth — focused on African languages

• TinyAya-Water — covering Asia Pacific, West Asia, and Europe

Why this matters

Most AI today assumes:

• Reliable internet

• Cloud access

• English-first usage

Tiny Aya breaks that assumption.

By running locally and supporting regional languages out of the box, Cohere is targeting governments, enterprises, and developers building AI for real populations, not just Silicon Valley demos.

It’s also a quiet challenge to closed models:
If useful AI can run offline, in local languages, on cheap hardware — the cloud monopoly starts to crack.

The bigger picture

As AI adoption spreads globally, the next billion users won’t come from English-speaking markets. They’ll come from regions where connectivity is patchy, devices are modest, and language diversity is non-negotiable.

Cohere is betting that open, efficient, multilingual models will win there.

Hot take

The future of AI won’t just be smarter.
It’ll be more local, more open, and more human-language first.

Tiny Aya isn’t chasing the top of the leaderboard — it’s chasing the world.