For years, global AI innovation has been dominated by the United States and China, while India — despite its massive talent pool — has rarely been viewed as a leader in core AI model development. That narrative is now shifting, thanks to Sarvam AI, a Bengaluru-based startup developing what it describes as a homegrown “sovereign AI.”

Sarvam AI is building foundational artificial intelligence systems from the ground up within India, and two of its latest products — Sarvam Vision and Bulbul — are drawing widespread attention for their performance.

Sarvam Vision, the company’s optical character recognition (OCR) model, is reportedly outperforming major AI platforms such as ChatGPT, Google Gemini, and Anthropic Claude in certain OCR benchmarks. Its accuracy and reliability have earned recognition from both industry professionals and everyday users.

According to posts shared by Sarvam AI co-founder Pratyush Kumar on X, Sarvam Vision achieved an 84.3 percent accuracy score on the olmOCR-Bench — surpassing Gemini 3 Pro and newer OCR solutions like DeepSeek OCR v2, while outperforming ChatGPT by a notable margin.

The model has also delivered strong results on OmniDocBench v1.5, a benchmark that measures how well AI systems interpret real-world documents. Sarvam Vision recorded a 93.28 percent overall score, excelling in processing complex layouts, technical tables, and mathematical content — areas where traditional OCR tools often struggle.

These achievements have boosted Sarvam’s reputation globally. The company, which was previously criticised for focusing on Indic-language AI models, is now receiving growing appreciation for addressing gaps overlooked by larger international AI firms.

Technology analyst Deedy Das recently acknowledged that he had underestimated Sarvam’s strategy, noting that the company has developed leading OCR, speech-to-text, and text-to-speech models for Indian languages at competitive pricing. Users have also expressed enthusiasm, praising the real-world usefulness of Sarvam’s tools.

Alongside its vision model, Sarvam AI has introduced Bulbul V3, a text-to-speech system designed to produce natural-sounding voice output in Indian languages. Comparable in concept to platforms like ElevenLabs, Bulbul V3 focuses on accurate pronunciation, expressive tone, and stability tailored for Indian-language applications.

Currently, Bulbul supports over 35 voice options across 11 Indian languages, with plans to expand coverage to 22 languages. The model has already gained adoption, with industry leaders such as KissanAI founder Pratik Desai highlighting its effectiveness and cost efficiency for Indic-language use cases.

With these developments, Sarvam AI is positioning India as a serious contender in the global artificial intelligence landscape.