India's Sarvam 105B LLM tops OCR benchmarks (timesofindia.indiatimes.com)

At the India AI Impact Summit 2026, Google CEO Sundar Pichai praised Sarvam AI for its groundbreaking work in developing local AI models tailored to Indian languages and contexts. Their latest model, Sarvam Vision, achieved an impressive 84.3% accuracy on the olmOCR-Bench, surpassing major players like Google’s Gemini and OpenAI’s ChatGPT. Sarvam AI is committed to unlocking India's vast pool of knowledge embedded in physical documents and historical collections, primarily focusing on enhancing AI functionality for Indian users by accurately processing 22 official Indian languages. The significance of Sarvam AI lies in its innovative approach to language and visual understanding, with capabilities that extend beyond standard text extraction to include image captioning, chart interpretation, and complex table parsing. Unlike global models that often neglect regional scripts, Sarvam is designed to provide higher accuracy for Indian languages, making it a vital tool for accurate information retrieval from diverse sources. Additionally, Sarvam’s speech recognition and translation models offer efficient processing with minimal device footprint, enabling bidirectional translations across multiple languages. A free Document Intelligence API will be available in February 2026, allowing developers to explore these advanced features and applications at scale.