Skip to content
Back to Blog
4 min read

India Unveils 12 Sovereign AI Models at Global Summit

At the India AI Impact Summit 2026, twelve Indian organisations are unveiling foundation models built for the country's 22 official languages.

  • India AI
  • sovereign AI
  • foundation models
  • Global South
  • multilingual AI

The India AI Impact Summit opened in New Delhi today, with heads of state, technology chief executives and policymakers from more than 100 countries. India is using it to unveil 12 indigenous foundation models built to serve 1.4 billion people across 22 official languages. It is the first major international AI summit hosted in the Global South.

His Highness Sheikh Khaled bin Mohamed bin Zayed Al Nahyan, Crown Prince of Abu Dhabi, is leading the UAE delegation. The UAE's participation centres on international cooperation in advanced technologies and strategic partnerships for AI that supports sustainable development.

The models

Under the IndiaAI Mission, twelve organisations are presenting foundation models trained on Indian data and built for regional languages. Four stand out.

Sarvam AI is developing a 120-billion-parameter open-source model focused on multilingual reasoning, speech technology and enterprise productivity. The government is expected to launch it formally during the summit, which could make it the largest open-source model built specifically for Indian languages.

BharatGen, a consortium led by IIT Bombay, is launching Param2, a 17-billion-parameter multilingual model with a mixture-of-experts architecture. It supports all 22 official languages and is trained on India-centric data gathered through the Bharat Data Sagar initiative. The consortium has also built speech models in 12 languages and document vision models.

Fractal Analytics is building what it calls "India's first large reasoning model", with 70 to 100 billion parameters and a focus on STEM and medical diagnostics. Its healthcare AI is expected to launch later this year.

Gnani AI is building voice-first foundation models for multilingual speech understanding and real-time voice interaction, aimed at customer service, education and accessibility.

Others are filling gaps around these. Gen Loop is developing lightweight language models in base, instruction-tuned and moderation versions across all 22 scheduled languages. Soket AI Labs is building open-source multilingual and multimodal systems for defence, healthcare and education. These teams are building from the ground up with Indian data and Indian languages as the primary focus, instead of fine-tuning Western models and hoping multilingual ability carries over.

Models already in use

BharatGen's models already run in production:

  • MahaGPT for the Maharashtra state government
  • regulatory AI assistants for the International Financial Services Centres Authority (IFSCA)
  • healthcare tools through the Medsum app
  • English assessment tools for the Kotak Education Foundation
  • policy explainers for citizen engagement

These systems serve citizens in languages that global foundation models handle poorly or not at all, and they give India's sovereign AI claims a credibility that benchmark scores alone would not.

Lessons for Arabic and the Gulf

The language problem India is tackling is familiar in the Arabic-speaking world, where most leading models are optimised for English first. Several parts of India's approach carry over.

Bharat Data Sagar shows the value of collecting and organising local data systematically for AI training. The UAE and Saudi Arabia have the same opportunity to build Arabic-centric datasets that capture regional variation in the language.

BharatGen shows how a consortium of leading engineering institutions can pool resources for foundation model work, which matters given how much capital these models need.

Voice deserves more weight. Several Indian teams, including Gnani AI and Gan AI, are putting speech first. Where literacy rates vary and oral traditions are strong, voice-based AI may reach more people than text.

The scale of the event

Organisers expect more than 250,000 visitors, 600 startups, 300 exhibitors from 30 countries and 13 country pavilions. OpenAI chief executive Sam Altman confirmed that India has 100 million weekly ChatGPT users, making it the platform's second-largest market. French President Emmanuel Macron and Brazilian President Lula da Silva are attending, as are Google's Sundar Pichai and Microsoft President Brad Smith. Seven working groups drawn from the Global North and South are developing proposals for shared computing resources and AI commons for public benefit.

India's AI market is projected to exceed $17 billion by 2027. The models unveiled this week are meant to keep more of that value inside the country instead of leaving it dependent on foreign AI providers.