Sarvam AI: Sarvam AI, a Bengaluru-based artificial intelligence startup, has managed to outsmart industry giants such as Google, ChatGPT and Anthropic Claude in understanding, reading and digitizing complex, sophisticated documents.
The company’s Sarvam Vision, an out-of-the-box artificial intelligence (AI) model, which was launched on February 5 claims to outperform many acclaimed AI systems such as ChatGPT and Gemini in terms of optical character recognition (OCR) and document intelligence for the languages of India.
Sarvam AI-What Makes It Different And Better?
Built around the proprietary vision-language model of the company, Sarvam Vision performed exceptionally well across 22 official languages of the country in OCR tasks (Malayalam, Kannada, Gujarati, Punjabi, Urdu, Hindi, Bengali, Tamil, Telugu, Marathi, Assamese and more).
Furthermore, Sarvam Vision is highly useful in its unique abilities to analyze complex visual structures. It can be used to interpret complex layouts, nested tables, multi-column documents and trend lines in charts. This unique capability makes it an excellent choice for examining academic papers, historical archives, financial documents and government records.
DON'T MISS
The company has decided to make its Vision Experience and Document Intelligence application programming interfaces (APIs) free in the month of February 2026 for all users.
The popularity and success of Sarvam AI would help India rely less on tools from other countries. This would also help in realizing the concept of Viksit Bharat 2047 while creating new jobs. In addition to these advantages, it would also help democratize access, which would eventually help developers and startups of the country to build India-centric applications. Moreover, introduction and success of tools such as Sarvam AI would foster indigenous innovation, reduce dependency on foreign tech and position the country as a world leader in inclusive AI for the world.
Sarvam AI Co-Founder’s Posts
Pratyush Kumar, the co-founder of Sarvam AI, recently shared intricate details of the in-house AI models’ achievements in a series of post on X (formerly Twitter). The shared posts suggested that an accuracy score of 84.3 percent was achieved by Sarvam Vision n the olmOCR-Bench. This score is higher than that of industry giants DeepSeek OCR v2, Gemini 3 Pro and ChatGOT.
Sarvam Vision also fared exceptionally well on the stringent benchmarks of OmniDocBench v1. The benchmarks evaluate the efficiency of artificial intelligence systems in reading as well as understanding documents in the real world. Sarvam Vision scored 93.28 percent overall, with astonishing results achieved in the context of mathematical formulas, technical tables and complex layouts.
