Maybe it’s me….but I find the quality of GPT 5 and 5.1 Codex to be progressively awful.
I have Gemini and Anthropic direct and also Cursor to compare to and I’ve pretty much stopped using GPT Codex.
Same question in the other solutions give great answers in <30 seconds, takes 5 minutes in Codex and the answer is usually vague and superficial.
I ask about code and it says ‘maybe this is happening’ instead of just tracing through the code.
Not usually one to rant - and I have no agenda but I thought for a change I’d put this up in the ‘Feedback’ forum just in case it makes any difference.
Maybe OpenAI has accepted defeat on the programming side of things now - for the last month or so my experience is that there is a very big gap. A gap that is now big enough for me just stop going to that toolset for dev.
Still good for API ‘narrative’ / ‘analysis’ type work but unreliable, slow and often of no value for coding.
I have to guess that more and more coding is being done using the other tools and maybe OpenAI has just accepted that GPT can’t be all things so better to go for the consumer revenue and cede the ground to the others for coding.
I wonder….
If a Codex window closes in VSCode and no one is there to listen does it make a sound?
Hmmm
GPT 5.1 Codex:
OpenAI’s own pricing page (https://platform.openai.com/docs/pricing) still lists GPT‑5.1 responses at $2.00 / 1K input tokens and $10.00 / 1K output tokens, which converts to $2,000 / 1M and $10,000 / 1M respectively. The cached‑input discount applies only to the cache-enabled SKUs (e.g., gpt-4o mini). GPT‑5.1 doesn’t currently support cached-input billing, so the $0.125 / 1M figure you saw on ChatGPT looks like hallucinated or outdated information. I’d trust the pricing page over the chat answer and keep the expert-mode cost multipliers aligned with the official per‑thousand numbers unless OpenAI publishes an update.
Confirmed. GPT-5.1-Codex-Max terrible performances even on Extra high. Furthermore, I found out that it is automatically switching to model such as GPT 4o even if in the selections says otherwise. GPT 5 Codex was mostly ok for me except the annoying “Stream disconnected before completion error”.