In the ongoing AI Impact Summit in New Delhi, India’s Sarvam AI launched two Large Language Models (LLMs), named Vikram, as an alternative to global AI systems dominated by Big Tech. The launch comes as the government-backed India AI Mission pushes for sovereign foundational models. Sarvam is expanding its base in hardware beyond models. This announcement has come months after the Centre announced that India would create its own large language model. They are not available to the public yet but the launch of the models — a 35-billion parameter model and a 105-billion parameter model — is being seen as a milestone in AI development in India. Both models have beaten comparable models from around the world, said Pratyush Kumar, co-founder of Sarvam. Google CEO Sundar Pichai also praised the work done by Sarvam AI.
What is Sarvam AI?
Sarvam AI was founded by Vivek Raghavan and Pratyush Kumar in August 2023. The indigenous AI model is capable of a range of visual understanding tasks, including image captioning, scene text recognition, chart interpretation, and complex table parsing. Sarvam AI aims to develop large language models (LLMs) fluent in Indian languages. It is trained on high-quality datasets covering 22 official Indian languages. Its speech recognition model supports 10 Indian languages within a single 74-million-parameter model that occupies approximately 294 MB on a device. It can recognize the language being spoken, without selecting it. The mean character error rate of the model is 0.0173 on a standard benchmark.
Difference between Sarvam AI and Gemini and Chat GPT
| Feature | Sarvam AI | Chat GPT | Gemini |
| Developer/Company | Sarvam AI (India) | Open AI (USA) | Google (USA) |
| Main focus | Indian languages, governance, local use-cases | General-purpose AI assistant | Deep integration with Google ecosystem |
| Languages supported | 22 Indian languages for its translation and document intelligence models, and 10–11 major languages supported for its conversational AI and voice models | 100+ global languages | 100+ languages |
| Best used for | Indian documents, OCR, speech, voice bots, governance AI | Writing, coding, research, reasoning, creativity | Gmail, Docs, YouTube, Maps, real-time search |
| Optical Character Recognition (OCR) and document reading | Excellent for Indian scripts | Very good | Very good |
| Speech and voice AI | Best for Indian accent voices | Very good | Very good |
| Coding | Limited | Excellent | Excellent |
| Image and video understanding | Basic/No | Advanced/Limited | Best |
| Best used for | Vernacular India use-cases | General-purpose AI | Video + Google workspace users |
Key Features
- Multimodal Vision-Language: Sarvam AI can interpret bith images and text altogether which helps read and analyze documents more clearly.
- Indian langauge focus: it provides high accuracy OCR and knowledge extraction in 22 Indian languages.
- Data interpretation: it understands charts, tables, diagrams, figures, illustrations and visual data inside documents.
- Multilingual visual: it can interpret many languages in the same document.
- Leading performance: it is one of the best AI systems for reading Indian language documents.
Sarvam AI is a crucial step towards achieving India-focused AI solutions, it is a step closer to India’s vision of Viksit Bharat 2047.
