India unveiled three homegrown AI models — one speaks 22 languages

India's IndiaAI Mission unveiled three homegrown foundation models in New Delhi on November 3: EKA-1 from Soket Labs and Dhīr 2B and Dhīr 5B from ZenteiQ, released by MeitY secretary S. Krishnan as a step toward what the government calls technological and strategic autonomy in AI.

EKA-1 is a 26.3-billion-parameter open sparse hybrid model trained on about 7.5 trillion tokens, built for math, code, reasoning and agentic work across all 22 scheduled Indian languages plus English, with banking, financial services and cybersecurity as its target sectors.

The Dhīr models are lightweight open-source assistants for English and 10 Indian languages that run on-premise and offline — even on mobile phones — for defence, government and low-connectivity settings, with weights released through the government's AIKosh platform.