ChatGPT 5.4 Pro Standard Mode - Adaptive Thinking or Nerfing Model?

I am using both Codex and direct REST API calls to GPT-5.4 Pro Standard and i don’t have the issues you are describing. But maybe we are served by different data centers. In my experience both Codex and API workflows benefit from evaluation, tooling, and better prompting practices that make your workflows more resilient to OpenAI’s backend changes. It is possible to build planning, issue tracking and knowledge DBs around the API. Are you referring to MathCo/Systemic? Would love to know more about their workflows and how they use the API. They surely use evals, no?