Blitz Bureau
NEW DELHI: Bengaluru‑based startup Sarvam AI has claimed that its latest vision and speech models have outperformed larger global rivals Google Gemini and ChatGPT on text‑to‑speech benchmarks for Indian languages. In a post on X, Sarvam AI’s co‑founder Pratyush Kumar said, “Sarvam Vision achieves state-of-the-art accuracy of 84.3 per cent on the OCR-Bench (English only subset) outperforming frontier models like Gemini 3 Pro and recent OCR models like DeepSeek OCR 2.”
On OmniDocBench v1.5 (English only subset), Sarvam Vision achieved 93.28 per cent overall score, excelling in complex formulas and layout parsing and being within touching distance of the current state of the art, Kumar added. Kumar also said the company’s Bulbul V3 text‑to‑speech model supports 35 voices across all 22 scheduled Indian languages and can handle different quality scans and content.
“On Indian languages, Sarvam Vision is the best model by far, while supporting all 22 scheduled Indian languages,” he claimed. The Vision series includes a 3‑billion‑parameter state‑space model capable of image captioning, scene text recognition, chart interpretation and complex table parsing.
Sarvam AI said its focus is on making artificial intelligence widely accessible to everyone in India. “We want India to embrace the most important technological shift of our time with confidence and control. Our ambition is to build foundational components and apply them to the country’s unique needs,” the AI company said. Kumar cited several examples on social media where the platform accurately extracted technical jargon from complex tables with merged rows and columns. Further, it showed Sarvam AI extracting data out of a chart from the latest Economic Survey.
