“ZAI's GLM-5.3 scores 66.9% on the DeepSWE benchmark, which is 0.5 points behind KIMI-K3, 3 points behind Fable, and 6 points behind 5.6 SOL.”