Gnani AI launches Artha sovereign AI stack for India

Artha is India’s clearest move yet to build its own stack for frontier grade AI rather than renting everything from US hyperscalers. An open weight 30B model tuned for 11+ Indian languages, wrapped in an agent platform and pushed as a sovereign stack for enterprises and the public sector, is exactly the kind of “national LLM” architecture governments have been sketching on whiteboards. The difference is that Gnani is shipping it with an explicit focus on cost per token and data residency. ([pib.gov.in](https://www.pib.gov.in/PressReleasePage.aspx?PRID=2304401&lang=2&reg=48&utm_source=openai))

For the race to AGI, the move does not suddenly put India at the bleeding edge of capability, but it does matter for who gets to shape the deployment context. If thousands of Indian banks, insurers and ministries build their AI workflows on top of Evon v3.3 rather than a closed US model, that creates a huge installed base of usage and domain data on a stack India can inspect and adapt. Over time, that can evolve into bigger, more capable models without surrendering control over safety, privacy and pricing.

It is also a signal to other emerging markets: you do not need a trillion parameter model to start asserting AI sovereignty. Mid sized, efficient models with strong localisation and open weights can be enough to anchor a regional ecosystem.