India's ambitious BharatGen AI model failed spectacularly in a recent voice transcription test by The Ken. The state-backed model, holding the largest share of India's AI Mission compute, struggled against venture-funded Sarvam and US-based Eleven Labs. This pits India's sovereign AI ambitions against the stark realities of execution and divergent funding structures.
How We Got Here
India's AI Mission backed 12 companies to build a "sovereign" AI ecosystem, with BharatGen and Sarvam designated as the two frontier models. BharatGen secured the mission's largest compute allocation to achieve this goal.
The Numbers
- BharatGen's Section 8 company structure prevents it from raising external venture capital funding.
- The Ken's engineers found BharatGen's model "took forever" to download from Hugging Face and required multiple installation attempts.
- Building an AI agent on Sarvam took an afternoon and cost Rs 45 per hour, while Eleven Labs was even faster and less than half the cost.
- In March, Maharashtra signed agreements with both models; by August, Sarvam's "Indus" assistant rolled out to 2,500 officials, while BharatGen's "MahaGPT" remains in pilot.
- Engineers who tested the models stated a clear preference for Eleven Labs, or Sarvam for rarer languages, over BharatGen even though it was free.
What Happens Next
🇮🇳 Why This Matters for India
For engineers and product managers building voice-enabled applications in Pune or Hyderabad, BharatGen’s practical difficulties mean significant delays and higher development costs, pushing them towards more robust alternatives.
The Take
India’s AI Mission risks consolidating frontier AI capability into one private player, Sarvam, given BharatGen’s structural and performance limitations. Policymakers must rapidly re-evaluate public-backed AI models to foster genuine competition, not just a single, state-supported champion.
Source:
The Ken ↗