Something significant happened in New Delhi today.
At the India AI Impact Summit, Sarvam AI quietly dropped two models that should be on every developer's radar: Sarvam-30B and Sarvam-105B. The 105B model doesn't just compete with global AI giants — it outperforms DeepSeek R1 on complex reasoning benchmarks.
India isn't waiting for permission to build world-class AI. It's already done it.
What Is Sarvam AI?
Sarvam AI is one of four government-selected sovereign AI builders in India — alongside Soket AI, Gnani AI, and Gan AI. These aren't hobby projects or research labs. They're part of India's national strategy to build AI that actually works for 1.4 billion people across 22+ languages.
Today's launch at the India AI Impact Summit is a milestone moment — not just for India, but for the global AI ecosystem.
The Two Models: What You Need to Know
Sarvam-30B — AI for Everyone, Including Feature Phones
The 30B model is built for accessibility at scale. It handles multilingual text generation, conversational AI, translation, and content creation — all in Indian languages including Hindi, Punjabi, Marathi, and 19 more.
The demo that turned heads: a feature phone holding a real-time AI conversation in Hindi. Sarvam's AI assistant "Vikram" — named in honour of Vikram Sarabhai — ran on it without breaking a sweat. That's not a gimmick. That's a signal about where AI is going: mass-market, multilingual, offline-capable.
Sarvam-105B — India's Answer to the Frontier
The 105B model is where things get serious:
- 105 billion parameters
- 128,000-token context window — handles entire documents, balance sheets, long-form reasoning
- Outperforms DeepSeek R1 on complex reasoning tasks (demonstrated live)
- Supports the same 22 Indian languages
The live demo: analysing a company's balance sheet in real time, responding in natural language. Enterprise-grade. India-built. Open source.
Why Developers Should Care
India just proved it can build frontier LLMs as well as anyone in the world. But the apps your clients actually need require more than a language model. They need image generation, video synthesis, voice and speech-to-text across 30+ languages, background removal, upscaling, and text-to-video.
This is where ModelsLab comes in. While Sarvam is solving India's LLM problem, we've been solving the everything-else problem — 10,000+ AI models across image, video, audio, and LLMs via a single API. 600+ agencies worldwide are already using ModelsLab to ship AI features in a day, not a month. No GPU management. No ML team. Just API calls.
Sarvam API Is Coming to ModelsLab
We're excited to announce that Sarvam models will be available via the ModelsLab API very soon. That means you'll be able to call Sarvam-30B and Sarvam-105B from the same API you already use for image generation, video creation, and voice synthesis. One API key. The entire AI stack. Built for India and beyond.
If you're building apps for Indian users — or global apps that need to speak Indian languages — this is the combination you've been waiting for.
Want early access? Drop us a line at sales@modelslab.com or book a demo below.
India's AI Moment Is Now
The narrative has shifted. India isn't just consuming AI built in San Francisco — it's building frontier AI that outperforms models from the US and China, in its own languages, at its own scale. For developers and agencies building for Indian clients, the question isn't whether to add AI features anymore. It's how fast you can ship them.
ModelsLab + Sarvam makes that answer: this afternoon.
Try the ModelsLab API: modelslab.com/playground
Book a 15-min demo: calendly.com/modelslab-sales/30min