India's Sarvam: 105B-parameter homegrown LLM shines in Hindi testing

abhish18 · x · 2026-08-27

The author traces Sarvam's story: selected under the India AI Mission, it launched a 105B-parameter model at February's AI Impact Summit, and he recently met founders Vivek Raghavan and Pratyush Kumar. In hands-on use, Hindi voice input transcribes with very few errors and is reportedly among the best across Indian languages — a made-in-India LLM worth trying.

Original post →

More from Models

Models channel →