AI-generated. This briefing is selected, summarised and published by our automated pipeline from the cited sources — no human editor reviews it before it goes live. How this site is made
The AI landscape is fracturing between compute haves and have-nots, as Meta explores renting its excess capacity while China's Kimi K3 matches Western models with a smaller team. This tension is mirrored by the release of both trillion-parameter models and 27B models that run efficiently on phones, creating a stark divergence in scale and resource requirements. The practical implications are already clear, with GPT-5.6 facing operational failures and specialized models like ZUNA1.1 demonstrating the value of focused development. For professionals, this signals a critical shift where the path forward depends on whether you are building for massive centralized infrastructure or efficient, accessible deployment.