Sarvam AI explained: What it is and how it outperforms Gemini and ChatGPT in India

Bengaluru-based startup Sarvam AI has emerged as one of India’s most closely watched artificial intelligence companies after its latest models delivered standout performance in India-centric AI tasks. With the launch of Sarvam Vision and Bulbul V3, the company claims it has surpassed global heavyweights such as Google Gemini and ChatGPT in specific benchmarks, particularly optical character recognition (OCR) and Indian-language text-to-speech.

Sarvam Vision tops global OCR benchmarks

Sarvam AI co-founder Pratyush Kumar revealed on February 5, via a post on X (formerly Twitter), that Sarvam Vision ranked first across major models on the olmOCR-Bench. This benchmark evaluates how accurately AI systems can extract and interpret text from images, scanned documents, and visual data, including handwritten content and complex fonts.

According to the benchmark results shared by the company, Sarvam Vision recorded an accuracy score of 84.3% on olmOCR-Bench, placing it ahead of Gemini 3 Pro and DeepSeek OCR v2. On OmniDocBench v1.5, the model achieved a score of 93.28%.

He also added, “On Indian languages, Sarvam Vision is the best model by far, while supporting all 22 scheduled Indian languages”.

Built for Indic scripts and Indian writing styles

One of Sarvam Vision’s biggest strengths lies in its deep focus on Indian languages and writing conventions. Unlike general-purpose AI models trained largely on global datasets, Sarvam Vision has been developed using Indian scripts and real-world document formats common across the country.

This targeted training allows it to better recognise regional scripts, mixed-language documents, and handwritten text in Indian contexts. While platforms such as ChatGPT and Gemini offer OCR capabilities, they are not fine-tuned specifically for Indic scripts in the way Sarvam Vision is.

As a result, Sarvam AI positions its model as a reliable solution for processing scanned documents, forms, and multilingual records, offering Indian businesses a homegrown alternative to overseas AI services.

At the core of Sarvam Vision is a 3-billion-parameter vision-language model designed for advanced visual understanding. The system can perform tasks such as image captioning, scene text recognition, chart analysis, and complex table extraction, making it suitable for enterprise document workflows and data-heavy applications.

Bulbul V3 raises the bar for Indian text-to-speech

Alongside Vision, Sarvam AI has also introduced Bulbul V3, a text-to-speech model that has attracted attention for its Indian voice generation capabilities. The tool has been rated ahead of ElevenLabs, a leading global voice AI platform, particularly in generating natural-sounding Indian voices.

Bulbul’s strong showing in India-focused benchmarks stems from its design, which closely reflects how languages are spoken across different regions of the country.

Bulbul V3 supports 35 distinct voices, spread across 22 official Indian languages. The voice samples span linguistic styles from the 1800s to the present day, accounting for variations in pronunciation, tone, and speech patterns across regions.

Kumar stated that Sarvam Vision currently offers the strongest support for Indian languages among available AI models, while covering the full range of officially recognised languages in the country.

Strong in niche areas, not a general-purpose rival

While Sarvam AI has delivered impressive results in OCR and Indian-language text-to-speech, its scope remains specialised. The company’s models are built for focused workloads rather than broad, everyday AI use.

Sarvam Vision’s 3-billion-parameter size is modest compared to general-purpose systems such as Google Gemini 3, which is rumoured to operate at nearly two trillion parameters. This gap highlights that Sarvam AI is not positioned as a direct replacement for ChatGPT or Gemini across all use cases.

Instead, its advantage lies in excelling at India-specific tasks where global models are less optimised, reinforcing its role as a purpose-built, sovereign AI solution for the Indian market.