IndiaAI Mission releases three home-grown models: Soket's EKA-1 and ZenteiQ's Dhīr 2B and 5B
At an event in New Delhi, the IndiaAI Mission unveiled three indigenous models built under its Innovation Centre pillar. Soket Labs' 26.3-billion-parameter EKA-1 supports English and all 22 scheduled languages, while ZenteiQ's compact Dhīr models are designed for secure, on-premise deployment and will be distributed through the AIKosh platform.
The IndiaAI Mission has released three foundation models developed by domestic companies selected under its Innovation Centre programme. MeitY Secretary S Krishnan presided over the launch in New Delhi, framing the models as evidence that subsidised national compute and research funding are starting to produce deployable technology rather than prototypes.
The largest of the three is EKA-1 from Soket Labs, a 26.3-billion-parameter model trained on roughly 7.5 trillion tokens. It is positioned for mathematics, coding, reasoning and agentic workloads, with a stated focus on banking, financial services and cybersecurity. Crucially for public-sector use, it supports English and all 22 scheduled Indian languages.
ZenteiQ AiTech Innovations contributed Dhīr 2B and Dhīr 5B, open-source multilingual models that cover English and ten Indian languages. Their selling point is footprint: both are small enough to run on GPUs, TPUs and even mobile devices, and have been built for secure, on-premise environments where internet connectivity is limited or disallowed.
The Mission provided the developers with subsidised access to the national AI compute facility and funding for ancillary research costs. Krishnan said the models will be made available to researchers, students and the public through AIKosh, the government's datasets and models repository, with intended use cases spanning document automation, information retrieval and citizen services across defence, government, industry and education.
Why it matters
Until now, India's sovereign-model narrative has been carried largely by a handful of well-funded startups. This release broadens the field, and the emphasis on small, offline-capable models is a pragmatic acknowledgement that most government deployments will run in constrained environments rather than hyperscale clouds.
Indicops view
Models such as Dhīr 2B and 5B are exactly the class we look to when a department needs language capability inside its own perimeter. The real work is not the model but what surrounds it: retrieval over departmental documents, evaluation in each target language, and audit trails that satisfy procurement and security reviewers. We expect to benchmark these models against existing Indic options for client workloads over the coming weeks.
More in Models
Back to the newsroomGoogle launches Gemini 4 Hydrogen, a $1.50-per-million-token model for coding and agents
Google has released Gemini 4 Hydrogen, a faster and cheaper tier of its Gemini 4 family aimed at everyday coding and agentic work. It costs $1.50 per million input tokens and $7.50 per million output tokens across a 1M-token window, and is rolling out through the Gemini API, AI Studio and consumer products.
Alibaba opens a two-week preview of Qwen4-Max with a 1M-token context, but no benchmarks or weights
Alibaba Cloud has opened a time-limited preview of Qwen4-Max, its next flagship model, through the Token Plan only. It offers a 1M-token context and up to 393,216 output tokens, but the documentation lists no pricing, benchmarks or downloadable weights, and the model retires on 25 November.
Anthropic releases Claude Fable 5.2 and Mythos 5.2, 30 per cent faster and 20 per cent cheaper
Anthropic has released Claude Fable 5.2 for general use and Mythos 5.2 for vetted cyber and life-sciences organisations. The company says output is up to 30 per cent faster than the 5.1 generation, at $8 per million input tokens and $40 per million output tokens, with lower rates of safeguard blocks.