It took The Ken three days, a high-end laptop, and little money to build a voice-transcription AI agent on India’s state-funded large language model BharatGen.
BharatGen is one of 12 companies that the government is backing to develop a “sovereign” AI ecosystem.
BharatGen took forever to download from Hugging FaceHugging FaceA website where AI companies and researchers publish their models for anyone to download and use , and several attempts to install.
Creating the same agent on Sarvam took an afternoon, an API key, a few lines of code, and about Rs 45 per hour.
The engineers who helped The Ken build the AI agent said they would always prefer using Eleven Labs—or Sarvam for rarer languages—over BharatGen, even though the latter was free.
It took The Ken three days, a high-end laptop, and little money to build a voice-transcription AI agent on India’s state-funded large language model BharatGen.
BharatGen is one of 12 companies that the government is backing to develop a “sovereign” AI ecosystem. But it is one of only two chasing frontier capabilities. The other is Sarvam.
The Ken tested both, coding over a weekend with two programmers to build an agent for various Indic languages and comparing them with a global benchmark, Scribe v2 from the US-based Eleven Labs.
BharatGen took forever to download from Hugging FaceHugging FaceA website where AI companies and researchers publish their models for anyone to download and use , and several attempts to install. When the agent built on it did work, it interpreted background noises as Kannada or Telugu or Urdu, repeatedly glitched in transcribing a Hinglish interview, and took double the time than its peers to produce any outcome.
Creating the same agent on Sarvam took an afternoon, an API key, a few lines of code, and about Rs 45 per hour. Eleven Labs had an even faster process at less than half the cost. And the results for both were much more accurate.
The gap is starker outside test environments. In March, for example, Maharashtra signed agreements with both Sarvam and BharatGen to deploy them in government functions. By August, Sarvam had rolled out Indus, a ChatGPT-style assistant for government and enterprise use, to 2,500 government officials in 22 languages. BharatGen’s AI assistant MahaGPT remains at pilot stage. Meanwhile, on 29 August, Karnataka’s IT minister met with Eleven Labs executives to explore the use of its voice AI across skilling, healthcare, and citizen services.
The engineers who helped The Ken build the AI agent said they would always prefer using Eleven Labs—or Sarvam for rarer languages—over BharatGen, even though the latter was free.
“The downloading, the debugging,” they explained, “it’s just not worth it right now.”
That’s not necessarily a verdict on quality, but an impression of how the competitors are structured.
