Meta trains 8B model to match Claude Opus 4.5

Meta researchers trained an 8B AI model to match Claude Opus 4.5's performance on complex enterprise workflows. The smaller model relies on a runtime harness for execution feedback, state trackers, and control-flow mechanisms, achieving frontier-level results without the high cost.