Bengaluru-based voice AI startup Gnani.ai has launched Prisma v2.5, a speech recognition model designed specifically for Indian languages and complex speech patterns, aiming to surpass competitors like Sarvam AI in accuracy and real-time application suitability.
- Supports 12 Indian languages with dialect and code-switching adaptation
- Claims higher accuracy and lower latency versus global competitors
- Plans expansion to Japan, the Philippines, and the Middle East
What happened
Gnani.ai unveiled Prisma v2.5 on June 17, 2026, a speech-to-text model that supports 12 Indian languages and is built to handle real-world complexities such as dialect variation, mid-sentence code-switching, and background noise. This new release leverages 14 million hours of proprietary Indic speech data for training and is accessible to enterprises through APIs.
The company claims Prisma v2.5 outperforms competitors including Sarvam AI, ElevenLabs, and Microsoft based on internal and third-party evaluations of word and character error rates, although specific benchmarks have not been publicly disclosed. The model aims to improve transcription for short utterances, numbers, alphanumeric sequences, and named entities, addressing common error sources across regulated industries.
Why it matters
Accurate transcription of Indian languages with their diverse dialects and multilingual patterns is a significant challenge for speech AI, which Gnani.ai seeks to solve through localized training and data center hosting in India. This localization enables lower latency and better robustness for real-time applications like customer service telephony compared to models hosted abroad.
The launch occurs amid a strategic push by Indian voice AI startups for 'sovereign AI' amid global export restrictions and security concerns affecting AI deployments. Gnani.ai's focus on enterprise reliability and government collaboration, alongside competitors such as Sarvam AI, highlights the increasing importance of domestically developed AI solutions integrated with real Indian usage conditions.
What to watch next
Gnani.ai plans to extend Prisma beyond India to markets in Japan, the Philippines, and the Middle East, seeking to grow the product’s global footprint. The company aims to capitalize on its integration with Indian data infrastructure, ongoing R&D, and recent $10 million Series B funding to accelerate growth and competitive positioning.
Ongoing industry developments around AI sovereignty, data security, and real-time performance will shape how enterprise clients select speech recognition solutions. Watching how Gnani.ai continues to benchmark against international providers and expand functional capabilities will be critical for assessing its leadership and adoption across sectors such as banking, insurance, and healthcare.