Global AI Model Wave: Efficiency Gains, Cost Cuts, and Chinese Disclosure Rules

Meta researchers have trained an 8B-parameter AI model that matches the performance of Claude Opus 4.5 at a fraction of the cost, signaling a shift toward high-efficiency, low-cost inference. The breakthrough leverages a "harness" runtime layer to provide execution feedback, allowing the model to handle complex, multi-hour enterprise workflows without relying on a massive context window. This democratization of frontier-level performance could disrupt the current "bigger is better"...