Technology
Weekly Technology: open-weight frontier models and sovereign AI stacks race past closed APIs
Two releases in the same week reframed the model-supply landscape. Tencent published a 770-billion-parameter open-weights system on August 30, and India's Gnani AI formally launched the Gnani Artha stack built around the 30-billion-parameter Evon 3.3 model and the Plexus agentic platform at Uprashtrapati Bhavan two days earlier. Together they show two parallel pressures on closed-API providers: capability is no longer a moat, and national buyers increasingly want self-hosted systems whose weights, training data and inference paths stay inside their jurisdictions.