Enterprise voice AI startup Gnani.ai has launched its latest model called Prisma v2.5. A speech-to-text (STT) model, Prisma spans across 12 languages and offers dialect variation, ambient noise, and natural code-switching which is directly baked into the training distribution. Users will now be able to access the model using application programming interfaces (APIs).

Gnani Prisma v2.5 is trained on 14 million hours of proprietary Indic speech. The model claims to close the transcription gaps such as short utterances, numerals, alphanumerics, named entities, and domain-specific vocabulary in critical industries like BFSI, insurance, healthcare, and several others. These are the categories where errors carry direct downstream consequences in compliance workflows, CRM logging, and agent assist applications.

“Most ASR models are built for ideal studio conditions. Indian calls happen over compressed network lines, in at least two languages inside a single sentence, in accents no studio corpus has ever captured. Gnani Prisma v2.5 is built for that reality,” said Ganesh Gopalan, cofounder and chief executive officer of Gnani.ai.

The firm has benchmarked the model on key metrics such as word error rate (WER) and character error rate (CER). The internal evaluations and third-party assessments indicate that Prisma outperforms both global and local speech-to-text providers such as Elevenlabs, Sarvam AI, and Microsoft, the company said. According to Gopalan, one of the early testers of the model is a retail client who has migrated to Prisma from a global STT model. 

Prisma also delivers significantly lower latency because it is hosted on local Indian data centres such as form providers like E2E Networks, Gopalan told Inc42. According to him, the reduced latency makes the model better suited for real-time applications such as telephony and other voice-based environments, where global models can sometimes struggle due to higher response times. 

Gopalan positions Prisma as a global model. And while the latest release is limited to Indian languages, the plan is to further launch it in countries like Japan, the Philippines, and the Middle East region. 

The new model follows the December 2025 launch of Vachana STT, a foundational, enterprise-grade Indic speech-to-text model trained on over one million hours of real-world voice data. Gnani.ai also launched its voice-first AI model Inya during the ‘India Impact AI Summit 2026’. 

In March, the startup raised$10 Mn (around ₹94 Cr) in its Series B funding round led by Aavishkaar Capital. The round also saw participation from existing backer InfoEdge Ventures. The startup is deploying the fresh capital to expand its customer base by exploring new verticals and global markets. Additionally, a chunk of the funds will go towards research and development and talent acquisition.

Gnani.ai, along with players such as Sarvam AI, Fractal Analytics, and BharatGen, are in the race to provide sovereign models in the highly competitive AI landscape. One of the most closely watched AI battles in India is unfolding in the voice technology space, where homegrown sovereign AI startups are increasingly competing with global players such as ElevenLabs and Wispr Flow. 

“Global companies are entering India because we are seeing early adoption of voice AI, and enterprises here are far more open to adopting this technology. However, we are fairly confident about our models. Because they are deeply localised and have been tested across a variety of real-world Indian environments, we believe they deliver higher accuracy and are better suited to the market than many competing models,” said Gopalan, when asked about the growing global competition. 

To be sure, Gnani.ai is one of the startups backed under the India government-backed IndiaAI Mission to build sovereign AI models tailored for India, along with the likes of Sarvam and Soket AI. 

A new case for sovereign AI is being hotly debated now, with the recent US restrictions on Anthropic’s most advanced AI models. Last week, the US imposed export controls on Anthropic’s Fable 5 and Mythos 5 models, citing national security concerns. The directive barred access by foreign nationals, prompting Anthropic to disable the models globally rather than selectively enforce the restrictions. 

The move has highlighted the risks of relying on frontier AI systems controlled by a single country and strengthened calls for sovereign, locally developed AI models in markets such as India and Europe. Gopalan, too, said that the episode highlights why countries need to develop their own AI models with strong security safeguards. 

[Edited by Nikhil Subramaniam] 

The post Gnani.ai Doubles Down On Sovereign Voice AI Models With Prisma v2.5 Launch appeared first on Inc42 Media.