India Launches 17B Multilingual AI Model
At the AI Impact Summit 2026 in India, developers showcased BharatGen Param 2, a new 17-billion-parameter multilingual model. The model is designed for deployable AI applications, particularly in large-scale urban systems and architecture. It represents a significant step in developing foundational models tailored for regional languages and use cases.
- The model utilizes a "Mixture-of-Experts" (MoE) architecture, which is a significant detail for builders. This means that instead of the entire 17-billion-parameter model being active for every task, only relevant "experts" within the network are engaged, leading to much greater efficiency and lower costs for inference. - The development of BharatGen Param 2 is a key part of India's "sovereign AI" strategy, which aims to create AI capabilities that are not dependent on foreign technology and are aligned with national interests. This approach prioritizes local data control and the development of an internal AI ecosystem. - In the field of architecture, AI is increasingly viewed as a collaborative partner rather than a replacement for human creativity. Tools are being developed to handle tasks like compliance checks and exploring countless design variations within parameters set by the architect, freeing up professionals for higher-level design thinking. - The model's focus on large-scale urban systems aligns with India's existing use of AI in urban planning. For instance, AI has been used in the "100 Smart Cities" mission to analyze traffic patterns and population growth to design more efficient road networks and green spaces. - For developers, the push for sovereign and more open AI models addresses a growing concern about the "silent erosion of developer sovereignty," where reliance on a few large, proprietary "black box" systems limits control and transparency. - The multilingual capabilities, supporting 22 Indian languages, are crucial for creative and technical work in a diverse country like India, enabling everything from generating culturally specific content to creating more accessible user interfaces for a wider audience. - The project is a collaborative effort led by IIT Bombay with nine other academic institutions and is backed by over ₹1,235 crore from the Indian government, indicating a significant national commitment to building foundational AI infrastructure. - Beyond architecture, the model is already being applied in healthcare through a partnership with Amrita Hospitals. A speech-to-text solution called Medsum AI is being used to help doctors generate clinical summaries, showcasing the model's real-world utility.