I was really frustrated by GPT-5.4, but last night I really pulled out the stops and within a few hours I got path tracing and DLSS implemented on top of Godot, which doesn’t even support DLSS. Just to see if it could do it? And you know what, it did, which was absolutely mind blowing. It wrote like 5,000 lines of C++, I set up a mostly local asset production pipeline using GPT image gen, voiceovers using ElevenLabs API, and even background music using Suno via the chrome use extensions in Codex. I just wanted to see how far I could push this little dumb game my kids asked me to make, and my kids are like “wow our game looks so good!” These models are absolutely mind blowing. I didn’t want to go to sleep I was having so much fun.

Adapt to what? If they are the "same", there is no reason to move. Actually, there are reasons not to, if you care about OpenAI's behavior.