Tech Mahindra's new Hindi AI model makes STEM learning personal
Tech Mahindra just dropped an 8-billion-parameter Hindi-first AI model focused on education, announced at the AI Impact Summit in Delhi.
Built under Project Indus and backed by the IndiaAI Mission, this new LLM aims to make learning in Hindi (and 37 dialects) smarter and more personal—especially for STEM subjects.
Model trained on synthetic tokens to address data gaps
The model uses NVIDIA's NeMo tools and was trained on 500 million synthetic tokens to handle data gaps.
It's designed to understand India's mix of languages and culture, so students get clear, relevant answers—even for tricky topics like physics.
The launch was announced on February 20, 2026.
Aiming for better, language-specific learning
While others like Fractal Analytics and IIT Bombay are building even larger or more general models, Tech Mahindra is zeroing in on making Hindi-based learning genuinely accessible.
If you're a student who prefers studying in your own language or dialect, this could be a big deal.