model = "gpt-5.4"
model_reasoning_effort = "xhigh"
service_tier = "fast"
model_context_window = 1050000

Codex피셜 실제로는 95퍼센트인 "model_context_window":997500, 로 들어간다고