No, no, just trust in the process! Better things are coming!
/s
I agree with you though. Better things can be coming and we can also all be angry that they ditched 5.3-Codex with little to no warning or fanfare and none of the currently available models come close to its efficiency or usefulness. Christ, if you’re going to take 5.3-Codex out back and shoot it, at least have a suitable replacement ready…
Chiming in to add my two cents - PLEASE bring back 5.3-codex. I blast through my 5 hour usage limit with 5.4-medium with like one or two tasks. I’m sure 5.5 consumes even more. It’s completely unusable with this consumption rate. With 5.3-codex I may hit my usage limit once in a while, but it was never really that impactful.
I’ve been using the Codex plugin for IntelliJ. For now I’ve switched over to Junie with Gemini 3 Flash. It’s doing at least well as 5.3, so far, and I can actually get work done with it.
So in short this consumption rate with 5.4 and 5.5 is pushing devs to use other models. OpenAI is not competitive.
Would like to chime in as well. I just signed up paid, was thinking I get to use 5.5 for planning and 5.3 for execution. This is no longer the case.
I hope any product managers listening brings this back - or - at least give us a model that’s dedicated to coding and producing code.
And only code.
I managed to keep access to 5.3 with printenv OPENAI_API_KEY | codex login --with-api-key, after creating an API key from my account. Might not work for all subscription types.
I guess the crowd feedback is pretty clear: 5.3-codex offered the right balance between reasoning, specification-conformance, speed and cost. 5.4 and 5.5 don’t.
Alexander Embiricos, you are probably responsible for a decision that will further accelerate subscription churn at OpenAI. Pulling the rug on the very things that made codex successful and enabled a community of developers to build full development processes with codex since 2021, is definitely not a smart move.
This will charge your API account directly and not use your ChatGPT subscription rate limited bundled allocation of course.
Curious how much 5.3 costs you this way. Please report back.
I tried this too. I didnt get 5.3 codex back. I tried 5.5 medium and racked up $10 in about 2 hours.
But I use chatgpt for planning, and then have it write a formatted prompt specifically for Codex in VScode with the assumption it has full IDE access.
I also tested 5.5 low but used planning mode first and spent about the same ($11) over 2 hours.
It’s absolute BS
With goal you cant even use 5.3 codex, in new updates. Now i am sing api and logged in via api
Not sure what to do any idea you people facing he same issue ?
$75 api wasted to fixing but still have a lot of bugs and glitches
I am officially out of OpenAI GPT, today for a simple logo change on 16 files (36 lines of code, eat 11% of the 5 hour limit on 5.4 medium reasoning). So long OpenAI, already got DeepSeek v4 PRO with 20$ credit for testing few days ago, coding all day it costed me a little less than 20 cents. Don’t even think returning any time soon, your new models are the worst by far for coding tasks running in circles for 20 minutes to change simply few lines of code, while the logo was provided by me not generated.
Yeah, don’t use /goal. It sucks!