India Unveils Gnani Artha Sovereign AI Stack

Overview

• The Gnani Artha stack was unveiled by Vice‑President C.P. Radhakrishnan on Friday, August 28, 2026, as a government‑backed offering for enterprises and public institutions.
• Artha combines Evon v3.3, a 30‑billion‑parameter large language model trained natively on 11 Indian languages, with Plexus, an agentic platform for building workflow‑specific AI agents.
• Gnani says Evon v3.3 was trained on about 2 trillion tokens, is available under an Apache 2.0 licence on Hugging Face on request, and claims about 40% token savings for Indic‑language workloads compared with comparable models.
• The company reports early commercial interest after showing Evon to select customers — five of 20 enterprises at a recent meeting began building on the model — but no Evon production deployments have gone live yet.
• Gnani plans larger 70B and 100B models, wider language coverage (from 11 to 22 languages) and speech‑to‑speech work, and the launch is being used to push India’s policy goal of indigenous AI for data sovereignty and local jobs.