Gnani AI Unveils Evon 3.3: 30-Billion Parameter Sovereign Open-Weight LLM
On August 28, 2026, Indian artificial intelligence firm Gnani AI officially unveiled Gnani Artha, an end-to-end sovereign enterprise stack featuring Evon 3.3, a 30-billion-parameter foundation model published under permissive open-weight terms on Hugging Face.
Evon 3.3 was developed through continual pre-training and domain-specific reinforcement learning on Nvidia accelerated computing clusters. The model specializes in complex reasoning across 22 scheduled Indian languages as well as English, addressing high-accuracy conversational processing in vernacular voice workflows, banking automation, and public sector governance systems.
The model is paired with Plexus, Gnani AI's proprietary agent orchestration engine designed to deploy low-latency conversational agents across edge telephony and on-premise data centers. According to benchmark disclosures reported by The Economic Times, Evon 3.3 demonstrates competitive domain accuracy in Indic financial comprehension and structured document processing compared to broader general-purpose open models like Qwen3.8-27B and Gemma 4-28B.
The release marks a significant milestone in national sovereign AI initiatives, offering organizations in South Asia an open-weight 30B foundation checkpoint that can be audited, fine-tuned, and deployed entirely within domestic regulatory boundaries without relying on overseas inference APIs.
