News · eGov
MeitY has launched EKA-1, Dhir 2B and Dhir 5B under the IndiaAI Mission's Foundation Model initiative, targeting offline mobile access and secure deployment in defence, banking and cybersecurity.
✦ AI SUMMARY 30-second read
- ZenteiQ's Dhir 2B and Dhir 5B can run directly on mobile phones without an internet connection, expanding AI access in low-connectivity areas.
- Socket AI's EKA-1 is a 26-billion-parameter model trained on 7.5 trillion tokens, designed for mathematics, coding and reasoning in air-gapped environments.
- MeitY Secretary S. Krishnan said the models could strengthen India's strategic autonomy in AI, particularly for enterprise use in secure, offline settings.
Why it matters: Indigenous AI models deployable offline and in air-gapped environments reduce India's dependence on foreign technology in sensitive sectors.
This summary was generated by AI. Editors may review and update it.
The Union Ministry of Electronics and Information Technology (MeitY) has launched three indigenous artificial intelligence models — EKA-1, Dhir 2B and Dhir 5B — under the Foundation Model initiative of the IndiaAI Mission. The models were developed by Bengaluru-based firms Socket AI and ZenteiQ.
The launch is aimed at reducing India's reliance on foreign AI technologies and building domestic capabilities suited to the country's specific requirements, including deployment in sectors where data security is a priority.
What was announced
ZenteiQ's Dhir 2B and Dhir 5B are compact models designed to operate directly on mobile phones without an internet connection. Users can download the models, after which configuration takes place automatically on the device itself, according to ZenteiQ Founder and Chief Scientist Professor Sashikumaar Ganesan.
The company is developing a free mobile application for both Android and iOS platforms. Model weights are already available through AIKosh, and ZenteiQ plans to release a 9-billion-parameter model in December for researchers and engineers working on specialised problems.
Socket AI's EKA-1 is a 26-billion-parameter model trained on 7.5 trillion tokens. It is built for mathematics, coding and reasoning tasks and can be deployed on private servers, private cloud systems and air-gapped environments. The company plans to make model weights openly available within a month, with early access provided through a waitlist.
What the developers say
Socket AI CEO Abhishek Upperwal said the model was built to remove dependence on foreign AI, particularly in sectors where control over data is critical, naming defence, cybersecurity, banking and finance as priority areas.
Socket AI Chief Science Officer and Co-Founder Mayank Singh said the entire data preparation, training and evaluation process for EKA-1 was carried out in India.
Government's position
Speaking at the launch, MeitY Secretary S. Krishnan said the indigenous models have the potential to strengthen India's strategic autonomy in artificial intelligence. He highlighted their relevance for enterprise applications in secure, air-gapped environments — systems that function without any connection to external networks.
The launch forms part of broader national efforts to develop AI technologies deployable across sectors while maintaining greater control over data and digital infrastructure.
What happens next
ZenteiQ is preparing a free mobile application and expects to launch its 9-billion-parameter model in December. Socket AI intends to open model weights to the public within a month following an initial waitlist phase.
Elets will continue to follow developments under the IndiaAI Mission's Foundation Model initiative.