The IndiaAI models released on 3 November include one 26.3 billion parameter system for Indian languages and two smaller open models built for offline use. The IT ministry announced them at an event in New Delhi, according to News On AIR.

Quick answer

  • Models: EKA-1 from Soket Labs, and Dhīr 2B and Dhīr 5B from ZenteiQ.
  • Released: 3 November 2026 under the IndiaAI Mission.
  • EKA-1: 26.3 billion parameters, about 7.5 trillion training tokens, all 22 scheduled languages plus English.
  • Dhīr: open-source, English plus 10 Indian languages, designed for low connectivity.
  • Access: the ministry says the AIKosh platform can make them available to researchers and the public.

What the IndiaAI models are

So, three models were released together. EKA-1 comes from Soket Labs, while Dhīr 2B and Dhīr 5B come from ZenteiQ. Both companies were chosen under the IndiaAI Innovation Centre pillar, and the government says subsidised national AI compute helped them build the models.

IT Secretary S Krishnan said the release “marks an important step towards building a strong and strategic AI ecosystem”. He also said that the models can help government, industry, research and education.

EKA-1: built for Indian languages and finance

EKA-1 has 26.3 billion parameters and was trained on about 7.5 trillion tokens. It also supports all 22 scheduled Indian languages along with English. Soket calls it an open, sparse hybrid language model for maths, coding, reasoning and agent-style tasks.

Soket’s chief executive, Abhishek Upperwal, said it is tailored to India’s banking, financial services and cybersecurity needs. However, the coverage does not state EKA-1’s licence, so it is unclear how freely developers can reuse it. One wire report adds that it can run some inference on a phone without mobile data, but the official summary does not confirm that.

Dhīr 2B and 5B: AI that can work offline

Next, the two Dhīr models are open-source and support English and 10 Indian languages. ZenteiQ says they are light enough to run on devices such as phones, so they can work without cloud connectivity. Because data can stay on the device, privacy also improves, according to the company.

Meanwhile, the ministry points to offline, document-grounded tasks. For example, a model could search a set of local documents and answer questions without sending anything online. That suits defence, government, industry and education, especially where networks are weak or data is sensitive.

What this means for readers in India

Most people will not download a model this week. Still, these releases matter for the apps you may use later. Bank helpdesks, government portals and school tools could use Indian language models that run closer to the user.

Offline models also help in places with patchy networks, because they need no steady connection. Then there is privacy: a model that runs on the device does not need to send your documents to a server. If you want to know your rights over personal data, our explainer on DPDP data rights covers the basics.

What is still unclear

However, several details are missing from the official summaries. The parameter counts for Dhīr 2B and 5B are not stated, and the 10 Indian languages are not named. Also, no consumer app or phone rollout has been announced, so treat this as groundwork for developers first.

FAQ

What are the IndiaAI models released in November 2026?

First, they are EKA-1 from Soket Labs and Dhīr 2B and Dhīr 5B from ZenteiQ. Then the IndiaAI Mission released them on 3 November 2026 at an event in New Delhi.

Which Indian languages does EKA-1 support?

EKA-1 supports all 22 scheduled Indian languages plus English. However, Dhīr supports English and 10 Indian languages, which the reports do not name.

Can I run these models on my phone?

The Dhīr models are designed to be light enough for phones and offline use, according to ZenteiQ. However, there is no ready-made consumer app yet, so use today is mainly for developers and organisations.