I’ve been using GLM 5.1 instead of GPT 5.4 for a few days now, and it’s working smoothly.