India's AI Mission Drops Three Homegrown Models, One With 26.3B Parameters

India's homegrown AI push got three new flagships on Tuesday. The IndiaAI Mission, run under the Ministry of Electronics and IT (MeitY), released EKA-1 from Soket Labs Technology and Research, plus Dhīr 2B and Dhīr 5B from ZenteiQ AiTech Innovations. IT Secretary S Krishnan and senior ministry and IndiaAI officials joined both companies' leadership at the unveiling.

EKA-1 is the heavyweight. It packs 26.3 billion parameters and was trained on roughly 7.5 trillion tokens, covering all 22 scheduled Indian languages alongside English. Soket calls it an open, sparse hybrid language model built for math, code, reasoning and agentic work — and it can handle some inferencing right on a smartphone, no mobile data or WiFi required.

Soket CEO Abhishek Upperwal said the model was shaped specifically for India's banking, financial services and cybersecurity landscape, aimed at sector-specific needs and the country's fast-moving digital ecosystem.

ZenteiQ's pair are the lighter, deployable kind. Dhīr 2B and Dhīr 5B are open-source models supporting English and 10 Indian languages, designed to run as secure, on-premise multilingual assistants. They handle offline-capable, document-grounded automation and knowledge retrieval for defence, government, industrial and educational use — especially in low-connectivity or sensitive environments.

ZenteiQ CEO Prof. Sashikumar said the models run across different computing platforms, including GPUs and TPUs, and their lightweight architecture allows on-device deployment on phones without leaning on the cloud. That keeps data on the device, which the company frames as a win for privacy, security and accessibility.

The government credited subsidised access to national AI compute infrastructure, plus support for ancillary research expenses, for getting the models built — its case study in how public AI plumbing speeds up homegrown innovation. Both firms sit among the teams picked under the IndiaAI Innovation Centre (Foundation Models) pillar. Krishnan said such models strengthen India's technological and strategic autonomy, opening productivity gains across government, industry, research and education, while platforms like AIKosh can put them in front of researchers, students and the wider public.