“ZAI's GLM-5.2 model is ranked number one on BridgeBench for reasoning, reportedly beating Fable-5 at one-tenth of the cost and achieving 300 tokens per second.”