Codex’s intelligence has been so severely downgraded that it’s practically unusable. I haven’t even been using it frequently—my weekly quota doesn’t reset for another three days, and I haven’t even used up 15% of it. My account is in good standing, yet it seems the requests are being routed to a model like GPT-4o-mini; it doesn’t “think” at all, and the accuracy rate is under 5%. Today I asked it to draw me a pelican riding a bicycle, and obviously it gave me a big duck egg instead of a pelican.
Switching to a more advanced model doesn’t solve the problem either; the output quality remains largely the same regardless of which model is used.
reynoldsabernathy,
I have seen similar behavior in everyday conversation as well. Regression upon regression. Do let me know if you find a solution here.
Same here, Really disappointed of OpenAI, this absolutely deliberately