India’s Sarvam AI Unveils 30B and 105B LLMs at India AI Impact Summit

Bengaluru-based AI startup Sarvam AI has unveiled two new large language models (LLMs) — a 30-billion-parameter model and a 105-billion-parameter model — both trained from scratch. The announcement was made at the India AI Impact Summit in Delhi, aligning with India’s broader efforts to build sovereign AI capabilities.

The company said both models are based on a mixture-of-experts (MoE) architecture, designed to enhance efficiency while maintaining strong performance across reasoning, programming, and tool-use tasks.

Sarvam 30B: Efficient MoE Architecture

Speaking at the event, cofounder Pratyush Kumar highlighted the company’s scaling strategy:

“Large language models, we have, of course, been building models across a 3-billion-parameter dense model, but it is important to scale up, and there are two other models that we trained and are talking about releasing today. One is Sarvam 30 Billion,” said Pratyush Kumar, cofounder of Sarvam.

Key Technical Highlights

• 30 billion total parameters

• Activates only 1 billion parameters per token

• 32,000-token context window

• Trained on 16 trillion tokens

“It is actually a mixture-of-experts model, we have a 30-billion-parameter model, but in generating every output token, it only activates 1 billion parameters,” Kumar said.

This design significantly reduces inference costs and improves reasoning efficiency.

“If you look at thinking budget, Sarvam 30B significantly outperforms both at the 8K and 16K scales compared to the latest models released at the same size.”

Kumar emphasised scalability as central to the company’s vision:

“We want to make AI work at population scale. Being able to do it efficiently becomes a very core thesis.”

Sarvam 105B: Positioned Against Global Frontier Models

Sarvam also introduced a 105-billion-parameter MoE model, activating 9 billion parameters per token and supporting a 128,000-token context window. It is designed for complex reasoning and agentic workflows.

“We trained a 105-billion-parameter model, it is also designed to do complex reasoning tasks very well,” Kumar said.

Benchmark Comparisons

Kumar compared the model against major global AI systems:

“At 105 billion parameters, on most benchmarks this model beats DeepSeek R1 released a year ago, which was a 600-billion-parameter model.”

DeepSeek’s DeepSeek R1, released last year, reportedly featured 600 billion parameters and was among the largest open reasoning-focused models at the time.

“It is cheaper than something like a Gemini Flash, but outperforms it in many benchmarks,” Kumar said.

Google’s Gemini Flash is designed as a lightweight, low-latency deployment model within the Gemini family.

On multilingual performance, Kumar added:

“Even with something like Gemini 2.5 Flash, which is a bigger and more expensive model, we find that the Indian language performance of this model is even better.”

India’s Sovereign AI Push Gains Momentum

The launch comes amid India’s accelerating efforts to develop domestic foundational AI models tailored to multilingual and large-scale public use cases.

The IndiaAI Mission, backed by a Rs 10,000 crore fund, aims to build India’s sovereign foundational AI model ecosystem.

So far, the mission has disbursed Rs 111 crore in GPU subsidies, with Sarvam AI emerging as the largest beneficiary.

The startup secured:

• 4,096 NVIDIA H100 SXM GPUs

• Allocated through Yotta Data Services

• Nearly Rs 99 crore in subsidies

Sarvam was previously selected as the first startup to build India’s foundational AI model under the mission.

Backing and Founding Team

Sarvam AI is backed by prominent investors including:

• Peak XV Partners

• Lightspeed Venture Partners

Founded in July 2023 by:

• Vivek Raghavan

• Pratyush Kumar

Both founders previously worked at AI4Bharat, supported by Infosys co-founder Nandan Nilekani.

The company offers a full-stack generative AI platform, ranging from research-driven model development to enterprise-grade deployment solutions.

Conclusion

Sarvam AI’s launch of the 30B and 105B models marks a significant milestone in India’s sovereign AI journey. By leveraging mixture-of-experts architecture for efficiency and claiming competitive benchmark performance against global players, the startup is positioning itself as a serious contender in the large language model space.

With government backing, GPU subsidies, and growing enterprise demand for multilingual AI systems, Sarvam AI appears set to play a central role in shaping India’s next phase of AI innovation.