I’ve been a ChatGPT Pro subscriber since around May, and until recently Codex had been working incredibly well for me.
My weekly usage would typically last about a week, resetting as expected. It was reliable, predictable, and honestly one of the main reasons I was happy to pay for Pro.
Over the last few days, however, everything changed.
The same type of work that used to last me an entire week now exhausts my usage in just one day. The newer models also seem to consume significantly more usage without any clear warning before a task starts.
What confuses me even more is the new pricing structure.
When I originally subscribed, Pro was marketed with a much higher usage allowance than Plus. Now there is a $200 tier offering that higher allowance, while the practical value of my existing subscription feels dramatically reduced.
My biggest concern is transparency.
If something is already working well, why change it?
If usage limits are changing, tell your customers clearly.
If new models consume more resources, warn us before we run a task.
Existing paying subscribers deserve to understand what changed and why.
I’m not asking for unlimited usage. I’m asking for transparency, predictability, and respect for customers who have been supporting the platform for months.
The $100 5x usage is actually new for this year, the $200 20x is the original version. If your like me I’d advise you to use Luna for small tasks and Terra for everything else. Sol is really a “break the glass” solution for extremely complex tasks in codex.
I’ve noticed something similar. My workflows haven’t changed much, but my usage seems to disappear much faster than it did a few months ago. I understand that newer models can be more resource-intensive, but it would really help if OpenAI provided clearer information about how different models affect usage before we start a task.
I don’t mind reasonable limits, but consistency and transparency are important. If the way usage is calculated has changed, or if certain models consume significantly more quota, subscribers should be informed. That would make it much easier to choose the right model for the job and avoid unexpected limits.
My weekly usage is being consumed much faster than before, but I have also experienced additional issues.
I had around 12–13% of my usage remaining, and it suddenly dropped to zero even though nobody was using my account and no tasks were running.
I then added $10 in credits. The entire amount was consumed by a single quick request involving a simple GET call using Luna medium. It was not a long or complex task, so consuming $10 for one request does not seem reasonable.
Same here OpenAi say that is going to imporove in the token consuption but my weekly is going so fast and i cannot use it after a day or so. Unbelivable.
It’s so bad now that it takes 1-2 days for the limit to reach the floor. It’s incredibly irritating because I used to rely solely on Codex, and now I have to supplement my work with others (Kimi/Claude) to get the same work done as before.
Have you looked into the API? You can pay for usage and keep going!
The limits on the ChatGPT consumer accounts are to give the most people access around the world and keep it fair. At the same time, if you need more usage, it’s available all in one place!
I know this reply was directed at Luis, but this is exactly what I’m trying to understand.
You’re recommending Terra as a replacement for GPT-5.4, but what is the recommended replacement for GPT-5.5 XThinking?
Terra is simply not equivalent for my workflow. I’m working on a very large production codebase with 250k+ lines of code across hundreds of files. On some tasks even Sol Ultra needs serious reasoning to understand all the dependencies and make the right changes.
If Sol Ultra can struggle with those tasks, Terra obviously can’t replace XThinking for me.
So is Sol XThinking supposed to be the actual replacement for GPT-5.5 XThinking?
This is the part that needs clarification. Recommending Terra to reduce usage makes sense for simpler tasks, but it doesn’t solve the problem for users who actually depended on the higher reasoning models for complex software development.
