Soket Labs & ZenteiQ Launch Indigenous AI Models

Two Indian AI startups, Soket Labs and ZenteiQ AiTech Innovations, have unveiled homegrown large language models (LLMs) under the IndiaAI Mission. At a New Delhi event, IT Secretary S Krishnan witnessed the introduction of Soket's EKA-1 model and ZenteiQ's Dhīr 2B and Dhīr 5B models. The EKA-1 model prioritizes mathematics, coding, reasoning, and agentic applications, while the Dhīr models concentrate on smaller, multilingual assistants suitable for environments with limited connectivity.

The IT ministry highlighted that Soket's 26.3 billion parameter EKA-1 model has been trained on 7.5 trillion tokens, enabling it to support all 22 scheduled Indian languages and English. Designed specifically for India's banking, financial services, and cybersecurity landscape, it addresses sector-specific requirements and digital ecosystem nuances.

ZenteiQ's Dhīr models are open-source and allow on-premise deployment, making them suitable for document-grounded automation, knowledge retrieval, and various government and educational applications. Operating across different computing platforms, including GPUs and TPUs, their lightweight architecture facilitates on-device deployment on mobile phones, preserving privacy, security, and accessibility.

IndiaAI Mission's support for subsidised AI compute and ancillary research expenses assisted in the development of these indigenous models. The partnership between startups and government subsidies has bolstered India's technological and strategic autonomy while enhancing productivity across sectors. Sarvam AI also joined the race with its open-weight decision model, Prism, which targets enterprise workflows such as customer support routing and AI-agent activity evaluation.

As decision models gain popularity for their token consumption efficiency and strict output control, Indian enterprises stand to benefit from this rapidly growing market, projected to reach $71 billion by 2030.