Soket Labs & ZenteiQ Launch Indigenous AI Models

Homegrown AI startups Soket Labs and ZenteiQ AiTech Innovations have rolled out new indigenously-developed large language models (LLMs) under the IndiaAI Mission. 

The models – Soket’s EKA-1 and ZenteiQ’s Dhīr 2B and Dhīr 5B – were unveiled at an event in New Delhi yesterday in the presence of IT secretary S Krishnan.

Both startups were selected under IndiaAI Mission to develop domestic foundation models. These models address different deployment needs. EKA-1 targets mathematics, coding, reasoning and agentic use cases, while Dhīr models focus on smaller multilingual assistants that can operate in environments with limited connectivity.

In an official statement, the IT ministry said that Soket’s 26.3 Bn parameter EKA-1 model has been trained on nearly 7.5 Tn tokens. Described as an open, sparse hybrid language model, EKA-1 supports all 22 scheduled Indian languages as well as English. It is designed for tasks spanning mathematics, code generation, reasoning and agentic applications.

“With the support of the IndiaAI Mission, we have developed an advanced AI model specifically tailored to India’s unique banking, financial services and cybersecurity landscape. It is designed to address sector-specific requirements, understand India’s evolving digital ecosystem, and deliver solutions aligned with the country’s needs and priorities,” said Soket Labs cofounder and CEO Abhishek Upperwal.

On the other hand, ZenteiQ also debuted Dhīr 2B and Dhīr 5B open-source multilingual models, which support English and 10 other Indian languages. As per the ministry, ZenteiQ’s AI models offer on-premise deployment and applications such as document-grounded automation, knowledge retrieval and use cases across government and education.

“Models like Dhīr 2B and Dhīr 5B offer the flexibility to run across different computing platforms, including GPUs and TPUs. Their lightweight architecture also enables on-device deployment on mobile phones, without dependence on cloud connectivity. This allows users to access AI capabilities locally while keeping data on their devices, enhancing privacy, security and accessibility,” said ZenteiQ CEO Sashikumar.

The announcement did not provide benchmark results, deployment details or pricing information.

Speaking at the event, Krishnan said that indigenous models will strengthen India’s technological and strategic autonomy while improving productivity across government, industry and research.

Both startups were selected under the mission’s IndiaAI Innovation Centre pillar to develop domestic foundation models. According to the government, subsidised AI compute and support for ancillary research expenses helped enable their development.

Sarvam Launches Decision Model

Separately, AI unicorn Sarvam AI also announced the launch of Prism, an open-weight decision model. Built on its Sarvam 105B model, the new offering will target enterprise workflows such as customer-support routing, security checks and evaluation of AI-agent activity.

The startup claims to have reported an average score of 71.8 across three evaluation indices and a median response time of 44 milliseconds on an NVIDIA B200 GPU. Sarvam said that Prism’s weights are available on Hugging Face and its own API, adding that local deployment will need a GPU with at least 120 GB of memory.

Sarvam’s new launch comes as decision models gain more traction among enterprises due to less consumption of tokens as well as strict output control and auditability. While the momentum began with Jev earlier this year, big giants like Cloudflare, Microsoft, AWS, OpenAI and Perplexity have also entered the space lately.

Homegrown startups are also catering to this shift. PurpleTalk’s Sthānika lab recently launched Sieve, while ConvAI Innovations has also unveiled an open-weight model called Laya that developers can run locally on laptops or phones. 

At the heart of all this is the Indian enterprise AI ecosystem, which is projected to become a $71 Bn market opportunity by 2030.

The post Soket Labs & ZenteiQ Launch Indigenous AI Models appeared first on Inc42 Media.