Sarvam AI has launched Sarvam Vision 2.1, an upgraded optical character recognition model tailored to Indian languages, addressing accuracy and cost challenges to better support businesses digitizing regional paperwork.
- Model improved with diverse real and synthetic data including handwritten Indic scripts
- Optimizations reduce costs for large-scale document digitization
- New Indic DiarBench benchmark developed for multi-speaker Indian language audio testing
What happened
Sarvam AI released Sarvam Vision 2.1, an enhanced version of its AI-powered document reading tool that converts scanned images into digital text using optical character recognition (OCR). This upgrade addresses previous limitations such as difficulty with complex forms, multi-page tables, and inaccurate text generation, known as hallucinations.
The company retrained the model using a combination of real-world samples and synthetic data, especially focusing on handwritten documents in multiple Indian languages. Additionally, the new version reduces computational costs, making it more affordable for enterprises to deploy at scale.
Why it matters
India's diverse linguistic landscape presents significant challenges for AI models, many of which perform well on English but struggle with Indic scripts. Sarvam's updates aim to create sovereign AI technologies that better serve local languages, supporting efforts to digitize a wide range of regional documents.
With AI deployment costs soaring globally, Sarvam Vision 2.1’s cost efficiency and benchmark-leading accuracy position it as a competitive option against international rivals, helping Indian businesses adopt AI-driven automation more economically and effectively.
What to watch next
Sarvam’s collaboration with IIT Madras’s AI4Bharat initiative led to the introduction of Indic DiarBench, a new evaluation suite for multi-speaker Indian language voice data. This benchmark fills a critical gap in voice AI resources beyond English and high-resource languages, signaling Sarvam’s broader commitment to Indian language AI development.
Looking forward, industry observers should monitor how Sarvam’s OCR and voice tools perform in real-world deployments at scale, particularly as Indian businesses increase efforts to digitize documents and processes across hundreds of regional languages and dialects.