Founded in 2019, Soket AI is tackling the challenge of creating advanced AI models for Indian languages and beyond by developing a 120 billion parameter open-source system designed to excel in reasoning, coding, and problem-solving across multiple underrepresented languages.
- Building a 120B parameter open-source AI model focusing on Indian and Global South languages
- Emphasizing reasoning, coding, and math skills alongside multilingual capabilities
- Targeting enterprise and government applications to reduce reliance on foreign AI providers
What happened
Soket AI, an Indian startup founded in 2019 by Abhishek Upperwal, is advancing a large-scale open-source AI model dedicated to Indian languages and other underrepresented languages from the Global South. The startup has received significant backing under the IndiaAI Mission, including ₹177 crore worth of funding and computing resources. Its flagship effort, Project EKA, aims to deploy a 120 billion parameter multilingual model by 2027.
Unlike many AI efforts that add support for regional languages as an afterthought, Soket AI integrates Indic languages as fundamental to its system. The startup also prioritizes advanced reasoning capabilities such as mathematics, coding, and logical problem-solving, aspiring to build a sovereign AI platform capable of serving diverse sectors like banking, cybersecurity, law, and government with minimal reliance on foreign AI technologies.
Why it matters
India’s linguistic diversity poses a unique challenge for AI, where dominant models remain heavily optimized for English and a limited number of global languages. Soket AI’s commitment to deeply supporting Indian languages fills a critical gap that many global AI providers overlook, advancing digital inclusion in one of the world’s most complex language markets.
Moreover, Soket AI's focus on reasoning functions such as coding from local language prompts represents a pioneering step that could democratize advanced AI tools beyond English speakers. This development is key for fostering homegrown AI technologies that support sensitive and strategic applications in India, potentially decreasing dependence on foreign AI suppliers and enhancing national data sovereignty.
What to watch next
Soket AI plans to launch its ambitious Project EKA multilingual AI model in 2027. Observers should monitor how effectively the startup balances advanced reasoning abilities with scale and affordability, especially given the technical challenges in data quality and infrastructure engineering.
Additionally, the model’s expansion beyond India to serve other underrepresented languages in the Global South will be critical for evaluating Soket AI’s impact on global language inclusivity in AI. How well the startup’s solutions integrate into enterprise and government systems and the extent of their adoption will signal the future competitiveness of indigenous Indian AI models on the world stage.