News · eGov
The IndiaAI Mission unveiled three indigenous AI foundation models in New Delhi, covering all 22 scheduled Indian languages and targeting sectors from banking to defence.
✦ AI SUMMARY 30-second read
- EKA-1, developed by Soket Labs, has 26.3 billion parameters and supports all 22 scheduled Indian languages plus English
- Dhīr 2B and Dhīr 5B, by ZenteiQ AiTech Innovations, are designed for secure, on-premise use in defence, government and education
- All three models are intended to be accessible to researchers, students and the public through platforms such as AIKosh
Why it matters: Indigenous foundation models supporting Indian languages could reduce reliance on foreign AI infrastructure across critical public and private sectors.
This summary was generated by AI. Editors may review and update it.
The IndiaAI Mission, operating under the Ministry of Electronics and Information Technology, unveiled three domestically developed foundation models in New Delhi on 3 November 2026. The models — EKA-1 and the Dhīr 2B and Dhīr 5B pair — were built by two Indian companies selected under the IndiaAI Innovation Centre's Foundation Models pillar.
IT Secretary S Krishnan, speaking at the event, said indigenous models strengthen India's technological and strategic autonomy in artificial intelligence.
What was announced
EKA-1, developed by Soket Labs Technology and Research, carries 26.3 billion parameters and was trained on approximately 7.5 trillion tokens. It supports all 22 scheduled Indian languages alongside English, with capabilities spanning mathematics, coding, reasoning and agentic applications. The model is tailored to banking, financial services and cybersecurity, and can perform some inference locally on devices such as smartphones without requiring mobile data or Wi-Fi.
Dhīr 2B and Dhīr 5B were developed by ZenteiQ AiTech Innovations. The open-source models support English and 10 Indian languages and are designed as secure, on-premise assistants. Their intended applications include document-grounded automation and knowledge retrieval for defence, government, industry and education, including in low-connectivity or sensitive environments. Like EKA-1, their lightweight architecture allows deployment on mobile phones without cloud connectivity.
Development support
Both developers were selected through the IndiaAI Innovation Centre's Foundation Models pillar. As part of that selection, they received access to subsidised national AI compute infrastructure along with support for research expenses.
The models are intended to be made available to researchers, students, innovators and the broader public through platforms such as AIKosh.
What it means for India
Foundation models are general-purpose AI systems trained on large datasets, making them reusable across a wide range of applications rather than limited to a single task. Their development within India, with explicit support for scheduled languages and offline deployment, could extend AI-driven services to regions with limited internet infrastructure.
The emphasis on on-premise and device-level inference, particularly for defence and government contexts, reflects a focus on data sovereignty alongside linguistic inclusion. Elets will continue to follow developments around the IndiaAI Mission and the deployment of these models.