Sarvam AI: 1 Trillion Parameter Model to Challenge OpenAI

AI startup Sarvam announces it is building a foundational AI model with over a trillion parameters. This model aims to challenge leading AI labs like OpenAI and Anthropic. The company made several announcements at its AI conference, Epoch 2026. These include the launch of Sarvam Vision 2.0 and Bulbul 4, alongside upgrades to its flagship Sarvam 105B model. Sarvam co-founder Pratyush Kumar states they are building this trillion-plus parameter model from scratch in India, to be competitive in areas like coding, cybersecurity, and simulation. The more parameters an AI model has, the stronger its performance. What's interesting is that the updated Sarvam 105B model is significantly cheaper than its competitors. It costs 5.5 times less than OpenAI’s GPT-5.4 Mini and 11 times less than Google’s Gemini 3.5 Flash for one million blended tokens. Sarvam also launched Vision 2.0, which focuses on advanced optical character recognition, especially for Indian handwritten text. This could offer cost-effective and powerful AI solutions for businesses.