GLM 5.2 is a 753-billion-parameter open-weights model challenging U.S. frontier pricing. Released by Beijing-based Z.ai, the mixture-of-experts model uses only 40 billion active parameters for high-speed inference while undercutting western API costs by up to 90%. However, lax enterprise token budgets and developer habits of defaulting to premium models currently protect incumbent U.S. labs from widespread migration.
Opening Kapyn…