India’s artificial intelligence ambitions have taken a major leap forward. Bengaluru-based startup Sarvam AI has announced that it is developing a trillion-plus parameter foundational AI model entirely from scratch in India. The company says the upcoming model will be built to compete with some of the world’s leading AI systems while significantly reducing inference costs for businesses.
The announcement comes at a time when AI companies worldwide are racing to build more powerful models without driving up operational expenses. Sarvam AI believes it has found a way to strike that balance, claiming its models are substantially cheaper than offerings from OpenAI and Google while maintaining competitive performance across enterprise workloads.
Sarvam AI’s Trillion-Plus Parameter Model
Speaking at the company’s launch event in Bengaluru, Pratyush Kumar, Co-founder of Sarvam AI, confirmed that the startup is developing a foundational model with more than one trillion parameters.
According to Kumar, the model is being built entirely in India and is designed to compete globally in areas such as coding, cybersecurity, scientific research, simulations, and enterprise AI applications.
Rather than adapting an existing open-source model, Sarvam AI says it is creating the system from the ground up, aiming to deliver world-class performance while keeping costs under control.
A Focus On Lower AI Costs
One of the biggest challenges facing enterprise AI adoption today is inference cost. Running large language models at scale can become expensive, particularly for businesses handling millions of requests every day.
Sarvam AI claims its flagship Sarvam 105B model is available at $0.80 per one million blended tokens. For comparison, the company says OpenAI’s GPT-5.4 Mini is priced at $4.50 per one million blended tokens, while Gemini 3.5 Flash costs around $9 for the same usage.
Based on these figures, Sarvam AI says its model is 5.5 times cheaper than GPT-5.4 Mini and more than 11 times cheaper than Gemini 3.5 Flash.
If these pricing claims hold up across large-scale deployments, they could make Sarvam AI an attractive option for enterprises looking to reduce AI infrastructure costs.
Built In India For Enterprise Workloads
Sarvam AI says its infrastructure is hosted entirely within India, an important factor for businesses and government organisations that require data residency and regulatory compliance. The company also revealed that it has already begun serving voice AI workloads, processing crores of voice minutes every day.
According to Sarvam, organisations building AI-powered voice products can benefit from lower costs, faster response times, and higher scalability compared to several existing global alternatives.
The company further claimed that competing AI models currently do not offer voice capabilities at the same operational scale.
Small Model, Big Performance
Despite having 105 billion parameters, Sarvam AI says its flagship model performs at a level comparable to significantly larger AI systems.
According to the company, the Sarvam 105B model delivers enterprise-level performance similar to GLM 5.2, a 753-billion-parameter model, across multiple benchmarks. These include:
- Tool usage
- Process execution
- Robustness
- Response consistency
If validated independently, this would demonstrate that efficient model architecture and optimisation can sometimes matter more than simply increasing parameter count.
Sarvam Vision Edge Also Debuts
Alongside its language models, the startup announced the commercial launch of Sarvam Vision Edge, a document intelligence platform designed for enterprise use. The platform focuses on extracting information from documents and streamlining large-scale document processing workflows.
Sarvam AI has also partnered with the Odisha Government on a research initiative aimed at digitising land records using its vision AI technology. The collaboration will use AI to identify, process, and extract information from government documents more efficiently.
Sarvam Vision 2.0 Enhances OCR
The company also unveiled Sarvam Vision 2.0, the latest version of its computer vision platform. According to Sarvam AI, the upgraded model significantly improves Optical Character Recognition (OCR) capabilities, making enterprise document processing faster and more accurate.
Improved OCR can help organisations automate workflows involving invoices, legal documents, government records, financial paperwork, and other large document repositories.
Conclusion
Sarvam AI’s announcement marks another significant milestone for India’s growing AI ecosystem. By developing a trillion-plus parameter foundational model from scratch, the company aims to compete directly with some of the biggest names in artificial intelligence while offering dramatically lower inference costs. Combined with new enterprise offerings like Sarvam Vision Edge and Vision 2.0, the startup is positioning itself as a strong contender in enterprise AI, voice intelligence, and document processing. While the trillion-parameter model is still under development, its progress will be closely watched as India seeks to establish itself as a global AI innovation hub.
Follow Us: Facebook | X | Instagram | YouTube | Pinterest
