IndiaAI Mission Unveils EKA-1 and Dhīr Foundation Models

The IndiaAI Mission, under the Ministry of Electronics and Information Technology (MeitY), unveiled three indigenous foundation models in New Delhi on 3 November 2026. EKA-1 was developed by Soket Labs Technology and Research; Dhīr 2B and Dhīr 5B were developed by ZenteiQ AiTech Innovations. IT Secretary S Krishnan said indigenous models strengthen India’s technological and strategic autonomy in artificial intelligence.

Capabilities and intended uses

EKA-1 has 26.3 billion parameters and was trained on about 7.5 trillion tokens. It supports all 22 scheduled Indian languages and English, and is designed for mathematics, coding, reasoning and agentic applications. It is tailored to banking, financial services and cybersecurity, and can perform some inference on local devices, including smartphones, without mobile data or Wi-Fi.

The open-source Dhīr 2B and Dhīr 5B support English and 10 Indian languages. Designed as secure, on-premise assistants, they provide document-grounded automation and knowledge retrieval for defence, government, industry and education, including low-connectivity or sensitive settings. Their lightweight architecture supports mobile-phone deployment without cloud connectivity.

Development support and access

Both developers were selected under the IndiaAI Innovation Centre (Foundation Models) pillar and received subsidised national AI compute infrastructure and support for research expenses. The models are intended to be available to researchers, students, innovators and the public through platforms such as AIKosh.

Important Facts for Exams

• Broad-data training: Foundation models are trained on extensive datasets so they can be applied across a wide range of use cases.
• Reusable model class: Foundation models function as reusable, general-purpose infrastructure, unlike bespoke models built for a single task.
• Transformer architecture: The Transformer architecture is the prevailing choice for building foundation models across different data modalities.