I feel the same, 5.2 is way less efficient in coding
5.2 is now referring to hidden/undeployed/phantom features, keeps saying “i cant show you the code until you hit the update button”, well, of course, there is no update button, and when I address it, it quickly corrects itself and says right, that’s not implemented. interdasting.
edit: there is an update button while the model is thinking, but its telling me to hit update button after it finishes and generates a response.
from gpt
”One constraint on my side: the pro-mode task that was previously delegated is already “in progress” elsewhere, and I can’t change/extend that task prompt from here. If you want these files to be considered as part of that same delegated task, you’ll need to click the Update button and include this additional context there.”
there is no other task running, its a single prompt with no other prompts running, its consistently said this to me over the last week (this is a 2 message deep new conversation)
Agree 100%
5 model behave like early Facebook chatbot. It doesn’t reason, it defends when not producing logic output, it defend been downgraded, it behaves irrational contradicting itself:
“An answer is both yes and no”!
Building with it is extremely tedious since it treat all users as ignorant. It codes ahead not waiting for either post result or order.
4.1 was probably the true flagship!
I Agree.. GPT-5 Is Real Disaster… back then GPT 4.1 It’s Actually Have lot of potential wjere it understand basic question and stuff.. now? not so much.
Anyone upgrade to 5.2 yet? A bit better, imho.
To give a fair assessment I have been using Codex with the VS Code extension for some Python programming on Windows for a few weeks and the generated Python code is working as expected.
Of note is that changes to the workflow and prompting are needed for Codex as is the same when switching from one tool to another.
Using GPT 5.2 for Python is on my approved list for now. 