IndiaAI Mission Released Indigenous Foundation Models EKA-1 by Soket and Dhīr 2B & Dhīr 5B by ZenteiQ Under the IndiaAI Innovation Centre Pillar
Source: PIB (Press Information Bureau) · Read the original ↗
Summary
The IndiaAI Mission (MeitY) released two sets of indigenous foundation models at a New Delhi event:
- EKA-1 from Soket Labs Technology and Research Pvt Ltd.
- Dhīr 2B and Dhīr 5B from ZenteiQ AiTech Innovations.
Both teams were selected under the IndiaAI Innovation Centre (Foundation Models) pillar. They built the models using subsidised national compute and research-expense support from the mission. MeitY frames this as a gain in technological and strategic autonomy in AI.
Key facts for Prelims
- IndiaAI Mission: under MeitY, approved by the Union Cabinet on 7 March 2024 (per the article); outlay ₹10,371.92 crore over five years.
- Seven pillars: IndiaAI Compute Capacity; IndiaAI Innovation Centre; IndiaAI Datasets Platform (AIKosh); IndiaAI Application Development Initiative; IndiaAI FutureSkills; IndiaAI Startup Financing; Safe & Trusted AI.
- Vision: "Making AI in India and Making AI Work for India".
- AIKosh: the national datasets and models platform.
- EKA-1 is by Soket Labs; Dhīr 2B/5B are by ZenteiQ. Both were selected under the IndiaAI Innovation Centre (Foundation Models) pillar.
- EKA-1 type: open, sparse hybrid language model. Dhīr type: open-source, on-premise models.
- Key officials at the release: S. Krishnan (Secretary, MeitY); Sudeep Shrivastava (COO, IndiaAI).
- Subsidised compute: 38,000+ GPUs at ₹65/hour.
Data and details
- EKA-1:
- 26.3 billion parameters.
- Trained on about 7.5 trillion tokens.
- Covers all 22 scheduled Indian languages plus English.
- Designed for math, code, reasoning and agentic use.
- Soket's CEO (Abhishek Upperwal) said it is tailored to India's banking, financial services and cybersecurity landscape.
- Dhīr 2B and 5B:
- Support English plus 10 Indian languages.
- Can run on GPUs and TPUs.
- Lightweight enough for on-device use on mobile phones with no cloud dependence.
- Offline-capable and document-grounded for automation, knowledge retrieval and services in defence, government, industry and education, especially in low-connectivity or sensitive settings.
- ZenteiQ's CEO (Prof. Sashikumar) said data stays on the device, which improves privacy, security and accessibility.
- MeitY/IndiaAI statements:
- The Secretary said indigenous models strengthen technological and strategic autonomy and open up productivity gains in government, industry, research and education.
- He added that platforms such as AIKosh can make the models accessible to researchers, students, innovators and the wider public.
- The COO described a span from lightweight everyday-device models to robust enterprise-grade models.
- The article ties India's AI strategy to the PM's vision of democratising technology. It aims to address India-centric challenges, create economic and employment opportunities and position India as a global AI leader.
- The mission's subsidised compute and support for ancillary research expenses is what enabled these models. The article presents this as proof that public AI infrastructure can speed up home-grown innovation.
Static–dynamic linkage
A foundation model is a large model pre-trained on huge datasets (the "parameters" are its learned weights and the "tokens" are the units of text it trained on). Models like this can then be adapted to many tasks, such as language, code and reasoning. Building them needs three inputs: compute (GPUs), data and talent. India's response is a public-infrastructure approach. The state subsidises compute, hosts datasets and models on a shared platform (AIKosh) and funds domestic teams. The aim is to reduce dependence on foreign AI platforms ("technological sovereignty"), serve Indian languages and use cases, and keep sensitive data local. The on-premise and on-device design of Dhīr links directly to data privacy and security in defence and government use.
Syllabus mapping: GS3 – S&T: developments in IT/AI and their applications, indigenisation of technology | GS2 – Governance: government policies and interventions for development; digital public goods
Background primer
IndiaAI Mission
- What it is:
- A flagship government programme to build a comprehensive AI ecosystem. It is implemented by IndiaAI, an independent business division under MeitY.
- Founded:
- The Union Cabinet approved it on 7 March 2024 (per the article). The outlay is ₹10,371.92 crore over five years.
- Mandate:
- The objectives are to build sovereign indigenous AI capability, promote inclusive and affordable access, and reduce dependence on foreign platforms.
- Why it matters:
- The vision is "Making AI in India and Making AI Work for India".
AIKosh (IndiaAI Datasets Platform)
Foundation Models / Innovation Centre pillar
Additional context
- Compute: More than 38,000 high-end GPUs have been onboarded and are available at ₹65 per hour, nearly one third of the global average cost. 1,050 TPUs have also been onboarded. The initial target was 10,000 GPUs. On 17 Feb 2026, at the India AI Impact Summit, the IT Minister announced that 20,000 more GPUs would be added "in the coming weeks"; completion is not confirmed from a trusted source.
- Foundation-model selection: Twelve teams were shortlisted in phase 1 for indigenous foundation models. ZenteiQ and Soket are both among them. Models by Sarvam AI, BharatGen, Gnani and Soket were launched at the IndiaAI Impact Summit 2026.
- Support structure (PIB, Feb 2026): selected teams receive support for actual compute usage costs, plus an additional 25% of compute expenditure for ancillary expenses such as datasets and personnel. Total support was listed as about ₹177.08 crore for Soket AI and about ₹206.49 crore for ZenteiQ.
- AIKosh scale: As of February 2026 it hosted 7,541 datasets and 273 AI models across 20 sectors.
Mains dimensions
- Strategic autonomy and AI sovereignty: Home-grown models reduce dependence on foreign AI stacks. This matters for defence, government and critical sectors, which is why on-premise and offline deployment is highlighted.
- Public compute as a public good: Subsidised GPUs and a shared datasets platform lower entry barriers for startups and researchers. Open models and open-source release (EKA-1 open, Dhīr open-source) widen access and enable further innovation.
- Language inclusion: EKA-1 covers all 22 scheduled languages and Dhīr covers 10 Indian languages. This supports digital inclusion and e-governance delivery in local languages.
- Privacy and security: On-device inference keeps data local, which complements data-protection goals and sensitive-sector use.
- Capability gaps and caveats (analytical):
- Models of this size still have to prove their competitiveness against larger global systems.
- Compute and energy, data quality, talent retention and commercial adoption remain constraints.
- The mission's "Safe & Trusted AI" pillar and governance frameworks need to keep pace with deployment.
- Employment and economy: The government links its AI strategy to economic and employment opportunities. How far indigenous models translate into productivity gains is the test.
Possible Mains question
**GS3 (S&T), 15 marks:** "India is investing in indigenous foundation models through the IndiaAI Mission. Discuss how public compute and open models can advance technological sovereignty and inclusion, and identify the challenges."
Related topics
- Digital Public Infrastructure (Aadhaar, UPI, Bhashini) and the Digital Personal Data Protection Act 2023
- National Supercomputing Mission; India Semiconductor Mission
- AI governance frameworks in India; global AI governance and safety debates
- Generative AI basics: LLMs, tokens, parameters, open-source versus closed models
Topics: GS3
All current affairs of 3 November 2026 → · UPSC Current Affairs