NEW DELHI: Tech Mahindra launches an 8-billion-parameter Hindi-first large language model on February 20. The company unveils the model at the India AI Impact Summit 2026. Tech Mahindra partners with NVIDIA under its Project Indus initiative.
The model provides AI-driven learning support in natural Hindi for foundational education subjects including physics. The architecture scales from Tech Mahindra’s previous 1.2-billion-parameter model. The upgrade reflects a significant expansion in computational capability.
The development team generated 500 million synthetic tokens using NVIDIA NeMo Data Designer. The synthetic training addresses data gaps in specific Indian languages. The model supports Agentic AI functionality for autonomous agents interacting in natural Hindi.
The partnership leverages NVIDIA NeMo framework for AI model development and NVIDIA NIM microservices for scalable deployment. The infrastructure ensures performance, scalability, and production readiness for educational applications.
“The model aims to democratize foundational education in subjects like physics for millions of students across India,” Tech Mahindra said in a statement.
Project Indus forms part of India’s broader push toward sovereign AI models. The initiative develops models with local languages, cultural context, and educational relevance. Tech Mahindra’s involvement reflects an industry drive to empower Indian learners with AI tools tailored to local needs.
The Hindi-first model is designed to reflect Indian linguistic and cultural context while supporting scalable artificial intelligence infrastructure. The collaboration used NVIDIA AI Enterprise solutions.
Previous
