Comment “model” and I’ll send you the links. While everyone was waiting for DeepSeek V4, India just released two open-weight reasoning models called Sarvam. An Indian startup trained 30B and 105B parameter models completely from scratch using government compute, not just fine-tuning existing models. Sarvam 30B is outperforming models like Gemma, Mistral, and Qwen on several benchmarks, while the 105B model competes with systems much larger than its active size. Both models also support all 22 Indian languages natively, making them one of the most interesting releases in the open-source AI space. #opensourceai #llm #sarvam #genai #aiindia