Gnani.ai launched Artha, an AI stack with a 30-billion-parameter model trained in 11 Indian languages. This "sovereign AI" play, endorsed by the Vice President, aims to cut India's reliance on foreign models. Early customer traction, however, shows only five out of twenty enterprises building on the new model.
How We Got Here
The IndiaAI Mission, launched in 2024, subsidises GPU compute for domestic models like Sarvam AI. Gnani.ai's new stack follows growing concerns about relying on US and Chinese AI infrastructure for sensitive data.
The Numbers
- Artha combines Evon v3.3, a 30-billion-parameter multilingual model, with Plexus, an AI agent platform.
- Evon v3.3 is an open-weight model available via Hugging Face under an Apache 2.0 licence.
- Vice President C.P. Radhakrishnan launched Artha at Upa-Rashtrapati Bhavan in New Delhi.
- Gnani.ai co-founder Ganesh Gopalan noted 5 of 20 enterprises in Pune expressed interest in Evon for underwriting and advertising.
- Artha aims to address challenges of data control, scaling costs, and effective operation across Indian languages.
What Happens Next
🇮🇳 Why This Matters for India
For government organisations and large enterprises in Delhi and Hyderabad, Artha offers a data-sovereign alternative for AI deployment, bypassing foreign infrastructure concerns.
The Take
The "sovereign AI" branding is compelling, but Gnani.ai's immediate challenge is proving the stack's reliability in regulated use cases like underwriting. This requires converting early interest into deployed agents, not just showcasing language capabilities.
Source:
MediaNama ↗