Three indigenous AI foundation models released under IndiaAI Mission – 'EKA-1', 'Dhīr 2B' and 'Dhīr 5B'

4 November 2026 · English Science & Technology

⬇ Download this article as free PDF

Topics: Science · 4 November 2026 Current Affairs (UPSC / TNPSC / SSC / RRB)

News

The Ministry of Electronics and IT (MeitY) released three indigenous AI models, built under the 'Foundation Models' initiative of the IndiaAI Mission, in New Delhi on 3 Nov 2026 in the presence of MeitY Secretary S. Krishnan. 'EKA-1', built by Soket AI Labs, has 26.3 billion parameters and was trained on about 7.5 trillion tokens; it supports all 22 scheduled languages and English. 'Dhīr 2B' and 'Dhīr 5B', built by ZenteiQ, are small, lightweight models supporting English and 10 Indian languages that can run on mobile phones without cloud connectivity. They can be used in data-sensitive sectors such as defence, cybersecurity and banking–finance, including in 'air-gapped' (offline, isolated) settings; the aim is to reduce dependence on foreign AI.

Key facts

Release	3 Nov 2026 – New Delhi; MeitY, IndiaAI Mission (Foundation Models pillar); Secretary S. Krishnan
EKA-1	Soket AI Labs – 26.3 billion parameters, ~7.5 trillion tokens; 22 scheduled languages + English; 'sparse hybrid' model; CEO Abhishek Upperwal
Dhīr 2B / 5B	ZenteiQ – open-source models; English + 10 Indian languages; run on mobile phones without cloud connectivity
Access	To be made available to researchers, students and the public via IndiaAI's 'AIKosh' platform
Use	Air-gapped / on-premise deployment – defence, cybersecurity, banking–finance, government

Background

• Foundation Models pillar: 12 organisations are funded under this IndiaAI pillar (Lok Sabha reply, Apr 2026) – Sarvam AI, Soket AI, Gnani AI, Gan AI, Avataar AI, BharatGen, GenLoop, ZenteiQ, Intellihealth, Shodh AI, Fractal Analytics and Tech Mahindra Maker's Lab. The largest grant went to the IIT Bombay-led 'BharatGen' consortium (₹1,058.52 crore). Sarvam's and BharatGen's models were unveiled at the 'India AI Impact Summit' in Feb 2026.
• Science & tech: a 'foundation model' is an AI model trained on vast data that can be adapted to many tasks. 'Parameters' are the internal weights learnt in training; a 'token' is the word/sub-word unit used in training.
• Security / sovereignty: an 'air-gapped' system is fully cut off from the internet and other outside networks, so sensitive data stays within the country. 'Open-weight' models can be downloaded and improved by researchers, students and firms. AIKosh is the IndiaAI Mission's national repository of datasets and AI models.

Tamil Nadu link: 'AI4Bharat', which builds free, open-source AI for Indian languages (machine translation, speech recognition, open datasets), is an IIT Madras initiative; the 'Nilekani Centre at AI4Bharat' was set up at IIT Madras with a ₹36 crore grant from Nandan and Rohini Nilekani.

Source: Analytics India Magazine, DefenceWatch, 3 Nov 2026; Indian Masterminds, Indian Television, 4 Nov 2026; MediaNama, 4 Apr 2026; Deccan Herald (AI4Bharat)

Join our UPSC classroom or online batch: UPSC coaching in Coimbatore · Fees · Call 99941 46662.

BANK Achievers