Bharatgen

India is moving to reduce its reliance on foreign AI systems as access to leading models becomes more exposed to geopolitics. In that shifting environment, IIT Bombay has introduced BharatGen, a domestic alternative built with India’s linguistic diversity at its center. The project is designed for a country of more than 1.4 billion people, where hundreds of languages and dialects make local-language AI more than a technical ambition. It is increasingly treated as a strategic need, especially as concerns grow that access to top-tier AI models may be restricted by the interests of the countries that develop them. A national push for AI independence BharatGen was unveiled at Bharat Innovates 2026 in Nice, France, as an open, multilingual family of AI models built specifically for India. IIT Bombay says the system is meant to understand and generate content across all 22 scheduled languages in the country. That focus separates it from many popular AI systems that have been built mainly around English and a small number of global languages. BharatGen is positioned as part of India’s broader effort to strengthen technological self-reliance while building AI that reflects local realities. The timing of the launch also matters. The global AI race is becoming more competitive and more political, with fresh concerns that access to advanced models could be limited by foreign policy and corporate decisions. More than a chatbot BharatGen is not being presented as a single conversational product. Instead, IIT Bombay has built it as a set of models for different tasks, each aimed at practical use cases in India. Model Primary Role Param2 Text foundation model for reasoning, programming, and tool use across India’s scheduled languages Shrutam2 Speech-to-text model for multiple Indian languages Sooktam2 Text-to-speech model with zero-shot voice cloning Patram Document and form understanding for workflows such as banking, insurance, and government services Param2 sits at the center of the ecosystem. The text-based foundation model is intended to support reasoning, coding, and external tool use while operating across India’s scheduled languages. Shrutam2 expands that reach into speech recognition, while Sooktam2 adds text-to-speech capabilities with zero-shot voice cloning. That means it can generate a voice similar to a speaker after hearing only a short sample. Built for public services and everyday workflows Patram gives the project another practical layer by focusing on the documents and forms commonly used in India. That makes it relevant for sectors such as banking, insurance, and government services. According to IIT Bombay, BharatGen has already been developed for use in government, healthcare, education, finance, insurance, and cultural preservation. The scope suggests the project is being treated as digital infrastructure rather than a narrow research experiment. The model family is also intended to support voice-first interaction in a country where comfort with text is not uniform across regions. In that sense, the project addresses language access and interface design at the same time. Backed by a national mission BharatGen is one of the most visible outcomes of the IndiaAI Mission, a roughly $1.2 billion initiative that gives selected startups and research groups access to subsidized compute. In return, the AI models they build are expected to be released publicly. The effort has been led by Prof. Ganesh Ramakrishnan of IIT Bombay’s Department of Computer Science and Engineering, alongside CEO Rishi Bal and Vice President of Machine Learning Dr Maneesh Singh. More than 60 researchers, engineers, and language experts have worked on the project within a consortium of nine top academic institutions. The involvement of language experts is central to the project’s design. Covering 22 scheduled languages requires not only computing power but also linguistic accuracy, especially if the models are expected to serve public-facing use cases. The launch also comes amid a wider debate about who will lead India’s AI future. One day earlier, Kris Gopalakrishnan, co-founder of Infosys, endorsed a social media post arguing that India’s large IT services firms have not produced a ChatGPT competitor because their business priorities are different. That argument points to a structural issue for the industry. Large public companies such as TCS, Infosys, Wipro, and HCL Technologies are built to deliver stable returns, create jobs, and earn foreign exchange, making long-term high-risk AI research harder to pursue at scale. As that debate continues, BharatGen places IIT Bombay and its academic network at the center of India’s generative AI conversation. The project’s significance lies not only in being built locally, but in its effort to make AI work in the languages that matter most to Indian users. Source: www.indiatoday.in

BharatGen Steps Into India’s AI Race, Built to Work Across 22 Local Languages