Skip to content

What Are EKA-1 and Dhīr? India’s New Homegrown AI Models

November 11, 2026

Tech & AI

By TrendScope Daily Team · November 11, 2026

Share WhatsApp Facebook X LinkedIn Copy Link

India does not have a ChatGPT or a Gemini of its own — not yet. But on November 3, the government took its most concrete step toward one, unveiling three homegrown AI foundation models built entirely by Indian startups under the government’s IndiaAI Mission. The models, named EKA-1 and Dhīr, are free, open-source, and designed from the ground up for India’s own languages and problems, rather than adapted afterward from a model trained mostly on English text from the West.

What was actually launched

At an event in New Delhi, S Krishnan, Secretary at the Ministry of Electronics and Information Technology (MeitY), unveiled three models built by two Indian startups selected under the IndiaAI Mission’s Innovation Centre pillar for foundation models. The first, EKA-1, comes from Soket Labs Technology and Research. The other two, Dhīr 2B and Dhīr 5B, come from ZenteiQ AiTech Innovations. All three are released as open-source models, free for researchers, startups and enterprises to use and build on through the government’s AIKosh platform.

EKA-1: built for Indian languages, math and money

EKA-1 is a 26.3-billion-parameter “sparse hybrid” model, trained on roughly 7.5 trillion tokens of data. Unlike most large language models, which bolt on Indian-language support as an afterthought, EKA-1 was trained from the start to work across all 22 languages in the Indian Constitution’s Eighth Schedule, plus English. Soket Labs built it with mathematics, coding, step-by-step reasoning and “agentic” tasks in mind — the kind of multi-step, tool-using work that AI companies everywhere are racing to get right. Soket’s cofounder and CEO, Abhishek Upperwal, has said the model is specifically aimed at India’s banking, financial services and insurance (BFSI) sector, along with cybersecurity — two areas where handing sensitive data to a foreign AI provider makes Indian regulators and boardrooms uneasy.

Dhīr 2B and 5B: small enough to run offline

Where EKA-1 aims for scale, ZenteiQ’s Dhīr models aim for reach. Dhīr 2B and 5B are mixture-of-experts models — a design that activates only the parts of the network needed for a given task, keeping them light enough to run on a laptop, a government server, or even a phone without an internet connection. They were trained from scratch on Google Cloud TPUs, support English plus ten Indian languages, and are built for document automation, multilingual assistance and knowledge tasks in defence, government offices, industry and education — settings where low-connectivity and strict data-security rules make a cloud-only chatbot a poor fit. ZenteiQ has released the model weights under the Apache 2.0 licence, along with technical documentation and evaluation tools, through AIKosh.

EKA-1 vs Dhīr 2B vs Dhīr 5B: how they compare

ModelBuilt bySizeLanguagesBuilt for
EKA-1Soket Labs26.3 billion parameters, trained on roughly 7.5 trillion tokensAll 22 scheduled Indian languages + EnglishMaths, coding, reasoning and agentic tasks; aimed at BFSI and cybersecurity
Dhīr 2BZenteiQ AiTech~2 billion parameters (per model name); mixture-of-expertsEnglish + 10 Indian languagesOn-device and offline use; document automation and multilingual assistants
Dhīr 5BZenteiQ AiTech~5 billion parameters (per model name); mixture-of-expertsEnglish + 10 Indian languagesThe larger of the two; defence, government, industry and education deployments
All three are open-source and distributed through the government’s AIKosh platform. Dhīr’s model weights are released under the Apache 2.0 licence.

“Make AI in India and Make AI Work for India.”

— IndiaAI Mission, stated aim

Why the government is doing this

The IndiaAI Mission was approved by the Union Cabinet in March 2024 with a budget of ₹10,371.92 crore (about $1.2 billion) over five years, meant to build a domestic AI ecosystem spanning computing infrastructure, startups, datasets and skilling — not just models. At the launch, Krishnan said the new models would strengthen India’s “strategic autonomy in artificial intelligence,” pointing to uses in mathematics, coding, scientific research and cybersecurity, including deployment in “air-gapped” environments where enterprise systems are deliberately cut off from the open internet for security reasons.

That framing matters because it’s the same argument driving sovereign-AI pushes in other countries: models trained mostly on English-language, Western internet data can miss context, idiom and nuance in Indian languages, and routing government or financial data through a foreign company’s servers raises its own data-sovereignty questions. EKA-1 and Dhīr don’t yet have published benchmark results against ChatGPT, Gemini or Anthropic’s Claude — officials have promised those “in the coming weeks” — so for now the honest answer to “how good are they?” is that nobody outside the labs that built them really knows.

What happens next

The rollout is phased: registered partners and pilot users get access first, with general availability targeted for early 2027. If the benchmarks hold up, EKA-1 and Dhīr would be India’s first real answer to the foreign-built models that dominate its AI app stores today — including OpenAI’s ChatGPT and Google’s Gemini, both of which have leaned hard into the Indian market this year. If they don’t, the government’s bet on home-grown foundation models will face the same question every sovereign-AI effort eventually does: is being made in India enough, or does it also have to be as good?

Sources: IndiaAI Mission launch coverage, Current Affairs Adda247, Swarajya Mag report on the EKA-1 and Dhīr launch

Quick questions

What is EKA-1?

EKA-1 is a 26.3-billion-parameter open-source AI model built by Soket Labs under India’s IndiaAI Mission. It supports all 22 scheduled Indian languages plus English and is designed for maths, coding, reasoning and agentic tasks, with a focus on banking and cybersecurity use cases.

What do Dhīr 2B and 5B do?

Dhīr 2B and 5B are smaller, open-source AI models from ZenteiQ AiTech Innovations, built to run on-premise or offline, including on mobile devices. They support English and ten Indian languages and target government, defence, industry and education use.

Are EKA-1 and Dhīr free to use?

Yes. Both are released as open-source models through the government’s AIKosh platform, with Dhīr’s weights licensed under Apache 2.0. Academic and public-sector users get subsidised access to the compute needed to run them at scale.

Related on TrendScope Daily


Share WhatsApp Facebook X LinkedIn Copy Link