Close

Specialised AI Models in India: Why Smaller AI Systems Could Be the Next Big Breakthrough

India is backing small, specialised AI models under the IndiaAI Mission, with 8 of 20 supported proposals being small language models for healthcare, languages and agentic AI. Cheaper, efficient and able to run offline, they could power AI at scale across the country.

Specialised AI Models in India

Specialised AI Models in India

For years, the AI race was about size, with bigger models promising better results. India is now backing a different idea: small, specialised models built for specific jobs, languages and devices.

What Are Small And Specialised AI Models?

Small language models (SLMs) have far fewer parameters than giant chatbots. They are trained for focused tasks such as healthcare queries, speech recognition or a single industry. That makes them cheaper to run and easier to deploy.

India’s Government Bets On SLMs

Under the IndiaAI Mission, the government has identified 20 indigenous AI model proposals, 12 large language models and 8 small language models. Union Minister Ashwini Vaishnaw told Parliament in July that domain-specific SLMs for healthcare, language technologies and agentic AI are at various stages of development and deployment. The Mission itself has an outlay of Rs 10,371.92 crore.

Efficiency Is Built Into The Design

India’s strategy talks about a “frugal, sovereign and scalable” AI ecosystem. Sarvam AI’s 105-billion-parameter model uses a mixture-of-experts design with roughly 10.3 billion active parameters. BharatGen’s Param2 is a 17-billion-parameter mixture-of-experts model covering 22 Indian languages. Both show that smart architecture can matter more than raw size.

AI That Runs Offline

Sarvam also showed Sarvam-Edge, a model that runs entirely on consumer devices without internet. For a country with uneven connectivity, that could bring AI to villages and small businesses.

Why Smaller Could Be Better For India

  • Lower cost: Startups and small firms can afford to run them.
  • Local languages: Models can be tuned for Indian languages and dialects.
  • Data privacy: On-device AI keeps sensitive health and financial data local.
  • Focused accuracy: A model built for one task can beat a general one on that task.

The Challenges Ahead

Experts note that India still has no frontier-scale model, and Sarvam’s 105B is competitive with mid-tier global models mainly on Indian-language tasks. Quality data in low-resource languages and sustained compute remain hurdles.

The Road Ahead

If India’s builders can pair affordability with accuracy, smaller AI systems could power healthcare, farming, education and governance at scale. The next breakthrough may not be bigger, but better targeted.

Aryan is a seasoned journalist with over 3 years of experience in digital media and news reporting. As the Editor at WEXT India, he brings sharp editorial insight and a passion for uncovering impactful stories in startups, business, and technology. With a strong foundation in journalism, Aryan ensures every article meets high standards of clarity, accuracy, and relevance.

Leave a Reply

Your email address will not be published. Required fields are marked *

Leave a comment
scroll to top