Technology
Archived — This article has been archived. The information may be outdated.
Gnani.ai launches 30-billion-parameter Indic AI stack
The Hindu BusinessLine··30 Aug
Bengaluru firm Gnani.ai launched Gnani Artha, a sovereign AI stack for Indian enterprises and public bodies, backed by the India AI Mission. The stack centres on Gnani Evon v3.3, a 30-billion-parameter open-weights model trained across 11 Indian languages, plus Gnani Plexus. CEO Ganesh Gopalan said larger models are planned later, including 70-billion and 100-plus billion parameter versions.
Prism
What It Means For You
- Enterprises can host the open-weights model on their own servers under the sovereign offering Gopalan described.
- Indic-language teams may see lower token use on Gujarati and Malayalam, which Gopalan said consume at least 40 percent fewer tokens than global models on the MILU benchmark.
- Public institutions evaluating India AI Mission-backed stacks now have Gnani Artha as a named option.
What's Happening
- Gopalan said Gnani started in 2017 with speech-to-text and calls Prisma its speech-to-text line for Indian languages.
- He said the firm is turning away some business because sales and support teams cannot expand as fast as demand.
- On regulation, he said India will sit between European and US approaches.
What Gnani Claims On Language
- Evon v3.3 is positioned for Indic language capability rather than as the firm's largest planned model.
- Gopalan said the company holds a large dedicated Indian-language speech dataset and continues daily data collection.
- He compared model choice to an Onam Sadhya, arguing different models serve different tastes and problems.
all-news
