When Sarvam AI—one of the first startups selected under the IndiaAI Mission—unveiled Sarvam-M, a 24B parameter hybrid LLM supporting 10 Indian languages, it aimed to be a landmark in sovereign AI. Instead, it stirred a storm. With just 334 downloads in two days, critics like Deedy Das of Menlo Ventures called it “embarrassing”, questioning the merit of “incremental work” and urging a deeper rethinking of India’s AI priorities.

When Sarvam AI—one of the first startups selected under the IndiaAI Mission—unveiled Sarvam-M, a 24B parameter hybrid LLM supporting 10 Indian languages, it aimed to be a landmark in sovereign AI. Instead, it stirred a storm. With just 334 downloads in two days, critics like Deedy Das of Menlo Ventures called it “embarrassing”, questioning the merit of “incremental work” and urging a deeper rethinking of India’s AI priorities.

Co-founder Vivek Raghavan called Sarvam-M an “important stepping stone” toward sovereign AI, while Aashay Sachdeva defended its performance, even posting proof of the model answering JEE Advanced 2025 questions in Hindi. Others, like @cneuralnetwork from AI4Bharat, highlighted the process over product: “The model is not the work—the process is.”

While critics compare Sarvam to OpenAI or DeepSeek AI, supporters argue it’s solving a completely different problem—building AI for Bharat, where hundreds of millions need tech in their native language. As Raj Dabre of Google aptly put it: “Before Sarvam-M came out, people were complaining about the lack of IndicLLMs. After Sarvam-M came out, people are still complaining.”

The debate exposes a bigger question: Can India’s homegrown AI efforts survive the hype-vs-impact divide? The answer might not lie in download counts, but in the voices of farmers, legal workers, and regional communities finally being heard—in their own language.

This post was originally shared by on Linkedin.