The IndiaAI Mission (MeitY) released two indigenous foundation models on 3 November 2026: EKA-1 by Soket Labs and the open-source Dhir 2B and Dhir 5B by ZenteiQ AiTech Innovations.
EKA-1 has 26.3 billion parameters, was trained on about 7.5 trillion tokens, and covers all 22 Scheduled Indian languages plus English; it is an open, sparse hybrid model for math, code, reasoning and agentic use.
Dhir 2B and Dhir 5B are lightweight multilingual assistants (English + 10 Indian languages) that run on-premise and offline, even on mobile phones, for defence, government, industry and education.
Both firms were selected under the mission's IndiaAI Innovation Centre (Foundation Models) pillar and used its subsidised national AI compute; the models can be shared through AIKosh.
'Making AI in India and Making AI Work for India' - build a sovereign AI ecosystem
Key: Approved by Union Cabinet on 7 March 2024; outlay Rs 10,371.92 crore over five years; implemented by the 'IndiaAI' Independent Business Division (IBD) under Digital India Corporation (MeitY); compute target of 10,000+ GPUs through public-private partnership
Cover the full AI value chain
Key: IndiaAI Compute Capacity, IndiaAI Innovation Centre, IndiaAI Datasets Platform (AIKosh), IndiaAI Application Development Initiative, IndiaAI FutureSkills, IndiaAI Startup Financing, Safe & Trusted AI
One national repository of non-personal datasets, models and tools
Key: Launched 6 March 2025 at the mission's first anniversary, alongside the AI Compute Portal
Fund Indian firms to train base models on mission compute
Key: Sarvam AI was first (April 2025, picked from 67 applicants); Soket AI, Gnani.ai and Gan.AI followed on 30 May 2025; eight more, including ZenteiQ, BharatGen, Tech Mahindra and Fractal, in September 2025 - 12 in all
Government-supported multilingual, multimodal LLM initiative
Key: Led by IIT Bombay under DST's National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS); its Param-1 model has 2.9 bn parameters trained on 7.5 trillion tokens; also selected under IndiaAI in September 2025
A large AI model trained on huge, broad data that can be adapted to many tasks. Parameters are its learned internal weights (model size); tokens are the word-pieces of training text (data size). A sparse model uses only part of its parameters for each input, cutting compute cost.
| Aspect | EKA-1 | Dhir 2B / 5B |
|---|---|---|
| Developer | Soket Labs Technology and Research Pvt Ltd | ZenteiQ AiTech Innovations |
| Size | 26.3 bn parameters, ~7.5 trillion tokens | Lightweight (2B and 5B variants) |
| Languages | All 22 Scheduled + English | English + 10 Indian languages |
| Design | Open, sparse hybrid; math, code, reasoning, agentic use | Open-source; on-premise, offline, on-device; document-grounded retrieval |
| Target use | Enterprise - BFSI, cybersecurity | Defence, government, industry, education in low-connectivity/sensitive settings |
GS Paper 3 > Science & Technology > Awareness in the fields of IT, Computers and AI
General Awareness > Science & Technology > Current Developments
The Constitution (71st Amendment) Act, 1992 amends the Eighth Schedule to the Constitution to include which of the following languages? 1. Konkani 2. Manipuri 3. Nepali 4. Maithili Select the correct answer using the code given below:
Answer: 1, 2 and 3
A large model pre-trained on broad data that can be adapted to many downstream tasks
IndiaAI's national datasets platform, launched 6 March 2025
A model that activates only a subset of its parameters for each input, lowering compute