Bengaluru-based artificial intelligence startup Sarvam AI has launched two new large language models, called Sarvam‑30B and Sarvam‑105B. The company announced the models on Wednesday at the India AI Impact Summit 2026 in New Delhi. These models are part of India’s effort to build its own advanced AI tools that can match global systems while handling Indian languages and local needs.
The new models are foundational AI systems, which means they are trained on large amounts of data and can be used for many tasks. These include chat conversations, coding support, research help, enterprise solutions, and more. These systems form a base that other applications can build on, similar to models like OpenAI’s GPT‑4.
The smaller model, Sarvam‑30B, has 30 billion parameters and a context window of 32,000 tokens. Sarvam says it was trained on about 16 trillion tokens and is built to run efficiently with less computing power. Benchmarks shared at the summit show that it performs well on tasks such as math reasoning, coding, and general problem solving when compared to many international models.
In demonstrations, the company showed a multilingual AI chatbot called “Vikram”. The chatbot could talk in languages such as Hindi, Punjabi, and Marathi. The team even showed it working on feature phones, not just smartphones, to reach users who do not have modern devices.
The Sarvam‑105B model is much larger, with 105 billion parameters and a 128,000‑token context window. It is designed for more complex reasoning and deeper tasks. At the summit, the company showed the model analyzing a company’s balance sheet in real time. The model answered detailed questions about the data. This shows the model can handle large documents, financial data, research work, and long business reports.
This launch is important for India’s AI industry. For the first time, an Indian startup has created large‑scale AI systems that claim performance close to international models and support many Indian languages. The focus on local language support, cost control, and data privacy may help Indian businesses and developers avoid relying entirely on overseas AI platforms.
Both models are planned to be open source, meaning developers and companies can use them to build their own AI tools. Sarvam says all the technology was built from scratch without depending on external data sources.
