Sarvam AI: 1 Trillion-Parameter Model & Vision Edge Launch

Indian AI startup Sarvam plans to build an artificial intelligence model with over one trillion parameters. This expands its AI infrastructure and model development. The announcement happened at the Epoch 2026 conference in Bengaluru. What's interesting is this new model is a significant jump from their current 30-billion and 105-billion-parameter models. Details on its architecture and availability are not yet public. Sarvam also launched Sarvam Vision Edge commercially. This document intelligence platform processes documents locally, extracting and organizing information from PDFs and scanned records. Edge-based processing reduces latency and helps with sensitive documents. The technology supports English and India's 22 scheduled languages. The company states its AI models are priced 5.5 times lower than global competitors, though this comparison is not independently verified. The bottom line is Sarvam is developing a full-stack AI platform, focusing on Indian languages and use cases.