Sarvam AI launches homegrown LLM suite for Indian languages
Sarvam AI just dropped its own large language model suite, headlined by Sarvam-105B—a 105-billion-parameter model that's outperformed DeepSeek R1 and Google's Gemini 2.5 Flash on Indian language benchmarks.
Vision model designed for OCR and document understanding
Trained on trillions of tokens, Sarvam-105B uses a smart mixture-of-experts setup for faster, more efficient results.
It performs competitively with big names like Gemma 27B and Mistral 32B in math, coding, and reasoning tests.
There's also a new vision model designed for OCR and document understanding across 22+ Indian languages, including Hindi and Tamil.
Pricing not disclosed yet
This entire AI suite was built in India using local talent, and it comes with handy extras like translation, text-to-speech, and agent systems.
Pricing has not been disclosed, but if you're into homegrown tech that actually delivers on performance (especially for Indian languages), this is one to watch!