Last week I was able to code for several days using the VS Code extention. I hit my weekly limit and was notified that I need to wait 3 days for my limit to reset. I searched around and found that Gosucoder had recommended Trae and Github Copilot as good “filler” subscriptions to make up the difference. I have been happily working in Trae but decided to switch back today.
I just checked my limits after a couple of hours and see that I’m at 95% for the week, but only 83% for the 5 hour period. I have ZERO recollection of using the OpenAI Codex weekly credits before this 5 hour session.
Trae says that they are providing my GPT-5 tokens, but is there any way that I could have been burning my OpenAI Codex credits while using Trae? I struggle to understand how else this could have happened (assuming the OAI token request counting is working properly).
Being able to see a histogram or some kind of daily count would really help.
There must have been a silent update to how Codex works, I have been using this no problem since release and today mysteriously I am now hitting usage limit reached and I didn’t even have that many tasks today. Time to start looking for an alternative to Codex I guess.
I just went to their help website and interacted with their AI chatbot. I sent screencaps showing that I hit 95% of my weekly limit in a single 5hr block of time. In that period, I think I had only about 2-3 requests. This should not happen logically. The chatbot agreed to escalate the problem to a human, so hopefully I’ll hear something soon.
I’m also using Trae for $10/mon and I really like it as a filler. Codex isn’t available, but I’m finding GPT-5 to be quite capable. I’ve not tried their other models but I will. In the past I was using Roo Code and it has also served me well but it isn’t as automatic in its agentic execution.
Thanks for the heads up, I’ll check out Trae. Codex used to be a no-brainer for me, but now that I am hitting these Balances very easily, it has turned me off their product. I don’t even recall seeing this usage dashboard until today, so this is definitely new. OpenAI is trying to milk us.
I’m on the pro plan, I’ve been using it extensively for the past 3 weeks or so, yet today, after not pushing off THAT many tasks, I hit my usage limit for the first time.
I wouldn’t mind it if I had visibility on what usage was actually being spent, but the dashboard wasn’t even working until now (I had glanced at it every now and then), and even then, the data on there is hardly useful for breaking down where the usage is being spent.
With their credit system, if I were to top up, that would be around $1 per cloud task per their estimates, yikes.
Yes, from what I see, you now need to become a pro user to be able to use it like a plus user before. I hope OpenAI rethinks this or optimizes their systems more in favor of Codex because, at the moment, Codex is ChatGPT’s selling point. Without it, I’m not so sure.
Same here. We used Codex (mainly web) quite successfully over last ~4 months. Quality is often surprisingly good. Yet, one of the key learnings is that we need 4 versions to get one that is of good quality.
Now, during the last 5h, I burned through 60% of my weekly quote (which came from ~5 tasks with 4 variations). The opaque credit system doesn’t really help either since it makes it impossible to predict what tasks costs.
I get that OpenAI needs to monetize but in its current form Codex simply isn’t a viable product…
Might be back to manual coding for now, until somebody cracks how to make AI coding viable.
Codex was what kept me from jumping to Claude Code. I understand it can’t be unlimited but piss poor communication and opaque limits that even their own backend seems to be unable to define with zero lead-time on changes…
What’s the point of making a model cheaper to run that Claude if you’re going to limit it as much as Claude? After all, Claude is better.
As soon as I hit the limit and got another shot to use it after the limit expired, I realized how worthless Codex is with limits. Suddenly I’m spending so much time curating my tasks for the LLM that I could do a lot of them myself instead, and better.
Yeah watching this usage drain on simple tasks now with now real output is embarrassing. OpenAI severely gimped Plus users, there is no point to use Codex if you’re on Plus now as far as it appears. There is no point in paying for Plus anymore at this rate. If you’re on Pro and maxing out, OpenAI’s silent update recently killed Codex. They even updated one of their articles related to token usage recently.
Shows the greed and it’s time to start seeking alternatives.
But nothing has been fixed? We received free credits that is great, it seems that daily usage is better now but it still burns through weekly usage after just a few prompts.
more strange things: my weekly hour limit should renew 11.14 but two hours past, it dropped from 86% to 84% while I sent 3 messages. How could this happen? How is it calculated???
@e1923 and @olicwyecontio, would you mind sharing more about hitting your weekly limit before hitting your 5h limit? This is definitely strange and maybe slightly different than the other issues we’re looking into, so we want to make sure we capture it correctly.
@OpenAI_Support this is the usage graph, weekly limit was hit on Nov 10/11 afaik with the limit resetting on Nov 12 11:34 PM
not once did I hit a 5h limit.. feel free to message me so I share the email address of the subscription so you can investigate in detail (it differs from the email address of this account)
