[image] 
之前使用5.3 xhigh类似的问题基本上没有超过20分钟。 
启动了fast模式，部分配置如下： 
model = "gpt-5.4"
model_reasoning_effort = "xhigh"
service_tier = "fast"

# ==========================================
# Codex 纯通用全局配置文件 (Base Global Config)
# ==========================================
model_context_window = 1000000
model_auto_compact_token_limit = 1000000

# OLD，暂存
approval_policy = "never"
sandbox_mode = "danger-full-a...