Sarvam AI Unveils Epoch Builder Platform for Multilingual Indian Language Models

Bengaluru-based startup Sarvam AI has officially launched its Epoch Builder Edition, a specialized platform designed to empower organizations to build, fine-tune, and deploy large language models (LLMs) tailored for Indian languages. Available in private preview starting August 2026, the platform aims to reduce reliance on foreign AI models by providing tools specifically optimized for the Indian linguistic landscape. General availability is scheduled for the fourth quarter of 2026.

The platform features new 7-billion and 70-billion parameter multilingual LLMs, which have been trained on a massive dataset of 2 trillion tokens. These models support over 10 Indian languages, including Hindi, Bengali, Tamil, Marathi, and Kannada. Beyond model access, the Epoch Builder Edition provides essential infrastructure, including GPU clusters for distributed training, reinforcement learning from human feedback (RLHF), and instruction tuning, alongside flexible on-premises deployment options for enterprise clients.

During the Epoch 2026 developer conference, Sarvam AI confirmed its commitment to building a foundational AI model with over one trillion parameters. This ambitious project is being trained entirely within India and is expected to go live within the next six months. By focusing on indigenous training data and localized infrastructure, Sarvam AI aims to set a new standard for sovereign AI development, ensuring that Indian enterprises have the necessary tools to innovate without depending on overseas technology stacks.