Gnani unveils sovereign AI stack ‘Artha’ featuring open-weight model, enterprise agents

Gnani’s launch of the Artha sovereign AI stack, featuring the Evon v3.3 LLM and Plexus agentic platform, marks a significant development for India’s domestic AI capabilities. The Evon v3.3 model, with its 30 billion parameters and Apache 2.0 open-source license, is specifically optimized for Indian languages. Its reported efficiency in token consumption, needing 20 percent fewer tokens per Indian-language word than GPT-5 family models, directly addresses the rising AI costs that enterprises face. This focus on cost-effectiveness and local language support positions Gnani to compete in the Indian market against global models. The emphasis on sovereign AI, allowing self-hosting and control over data, is critical for Indian banks, insurers, and government bodies navigating data residency regulations. Vice-President C P Radhakrishnan’s presence at the launch underscores the government’s push for open, affordable, and accessible AI built within India. The Plexus platform’s ability to deploy AI agents through natural language prompts, as demonstrated by fetching PAN card details or managing welfare programs, shows practical applications for enterprise and public sector use. The key challenge for Gnani will be proving Evon v3.3’s real-world performance and scalability against established global and other Indic models beyond benchmark tests like MILU. While the tokeniser efficiency is a strong selling point, adoption will hinge on consistent, high-quality output and ease of integration for diverse enterprise needs across India.