Gnani.ai just launched Prisma v2.5, an Indic speech-to-text model trained on 14 million hours of proprietary data. The Bengaluru startup claims Prisma v2.5 outperforms Sarvam AI, ElevenLabs, and Microsoft on accuracy, especially for real-world Indian calls. This pits home-grown AI startups directly against global giants in India's complex voice market.
How We Got Here
Gnani.ai raised $10 million in Series B funding in March, supported by Aavishkaar Capital and InfoEdge Ventures. The company previously launched Vachana STT in December 2025 and Inya at the India Impact AI Summit in 2026, building its voice AI portfolio.
The Numbers
- Prisma v2.5 supports 12 Indian languages, accounting for dialect, background noise, and code-switching directly in its training data.
- The model specifically improves accuracy for short utterances, numbers, alphanumeric strings, and named entities, addressing critical compliance and CRM issues.
- Gnani.ai cites lower latency by hosting Prisma in Indian data centres, including E2E Networks' infrastructure, optimizing it for real-time telephony.
- An unnamed retail client reportedly switched from a global STT provider to Prisma v2.5, citing performance gains.
- The company plans to expand Prisma v2.5 to Japan, the Philippines, and the Middle East, aiming for a global product.
What Happens Next
🇮🇳 Why This Matters for India
For call centre operators in cities like Chennai and Kolkata, improved Indic voice accuracy directly reduces compliance issues and customer service friction with diverse accents.
The Take
Gnani.ai's performance claims against Sarvam AI and Microsoft are bold, but the lack of disclosed benchmarks makes them difficult to trust. The winner in India's sovereign AI race will be the one that provides auditable, transparent results, not just claims.
Source:
MediaNama ↗