89
/
4
Min Read
Under the Central Government’s IndiaAI Mission, three indigenous AI foundation models EKA-1, Dhir 2B & Dhir 5B have been launched for public use. Developed by domestic companies Soket Labs Technology and Research and ZenteiQ AiTech Innovations, these models aim to expand access to AI in Indian languages and promote its applications across education, higher education, research, cybersecurity and industries.
The three AI models were officially launched by S. Krishnan, Secretary, Ministry of Electronics and Information Technology (MeitY), at an event held in New Delhi on November 3, 2026. Speaking on the occasion, he said that making indigenous foundation models publicly available is an important step towards building a strong and strategically independent AI ecosystem in the country. He added that these models could help improve productivity and operational efficiency across various sectors, including cybersecurity.
EKA-1: Supports 22 Indian Languages
Developed by Soket Labs, EKA-1 is a 26.3-billion-parameter foundation model trained on approximately 7.5 trillion tokens. One of its key features is its ability to support all 22 scheduled Indian languages, along with English. It has been developed as an open and sparse hybrid language model.
The model is designed to handle tasks involving mathematics, coding, reasoning and agent-based operations. Its multilingual capabilities could help students access AI-powered assistance in their preferred Indian languages. For higher education institutions, universities and researchers, the model could also prove useful for coding, technical subjects and academic research-related activities.
According to Abhishek Upparwal, CEO of Soket Labs, the company developed the model with support from the IndiaAI Mission, keeping in mind the requirements of India's banking, financial services and cybersecurity sectors.
Dhir 2B and Dhir 5B: Can Operate Without an Internet Connection
Developed by ZenteiQ AiTech Innovations, Dhir 2B and Dhir 5B are open-source, multilingual AI models. They support English and 10 Indian languages. One of their key features is their ability to operate without an internet connection on laptops, mobile phones and other devices.
According to the company, Dhir 2B has approximately 2.03 billion parameters, while Dhir 5B has 5.09 billion parameters. These models have been designed to support tasks such as extracting information from documents, automation and various digital services.
In education and research, these models could be particularly useful for institutions with limited internet connectivity or those that prioritise data privacy. Universities, colleges and research institutions could use them to analyse academic documents, retrieve information from study materials and simplify administrative processes.
According to Professor Shashikumar, CEO of ZenteiQ, the models can operate across different computing platforms, including GPUs and TPUs. Their lightweight architecture also enables deployment on mobile devices. This could help institutions process and secure data locally without transferring it to cloud platforms.
Technical and Financial Support Under the IndiaAI Mission
All three models have been developed under the Innovation Centre (Foundation Models) initiative of the IndiaAI Mission. Both companies were selected to develop indigenous AI capabilities and large language models. Under the initiative, they received access to national AI computing infrastructure at subsidised rates, along with support for research-related expenses.
Notably, the Central Government approved the IndiaAI Mission on March 7, 2024 with the objective of strengthening AI research, innovation, skill development and technological capabilities across the country.
The launch of these three indigenous AI models is expected to create new opportunities for developing AI-powered educational resources in Indian languages, promoting digital learning and strengthening research activities. Their offline capabilities and local data-processing features could also benefit educational institutions, particularly those operating with limited internet connectivity or requiring greater data security.
However, their actual impact will depend on their performance, reliability and practical adoption across different sectors.
