I still cannot understand what is happening with Codex. The limits are being burned at an absurd rate.
Before the GPT-5.6 series, a $20 subscription let me work with GPT-5.5 for weeks without constantly thinking about usage. I did not even know what it felt like to hit the weekly limit. I simply never got there.
I also have measurements from a particularly busy week before GPT-5.6. I still never hit the 5-hour limit, and the weekly limit was not even close. There was roughly 30-40% left.
I went back through my chat history and found screenshots I took on June 11, 2026:
-
10:20: before the weekly reset, 89% of the 5-hour limit and 47% of the weekly limit remained.
-
10:25: the weekly limit reset.
-
10:54: after the reset, 37% of the 5-hour limit and 94% of the weekly limit remained.
The weekly counter reset to 100%, while the 5-hour counter did not. Since the weekly limit was about to reset anyway, I decided to use GPT-5.5 Extra High to implement a large feature. It created 48 files, wrote about 1,600 lines of code, and worked for 40 minutes. The whole task used 8.3% of the weekly limit and 52% of the 5-hour limit.
And now? Two GPT-5.5 Medium prompts where I described exactly what needed to change and where. Six minutes of model work in total, including build time, and 180 lines of code. Result: 4% of the weekly limit gone.
I will not even get into Sol Low and Medium. This morning I gave it a task to check an already completed solution against the requirements. It worked for 23 minutes, added 130 lines, deleted one file, and used one Terra Low subagent as a code formatter. Around 14% of the weekly limit disappeared. This is not usable.
Over the last three months with Codex, I have never used an effort level above High on any model except Luna, where I sometimes used xHigh or Max. Most of the time I use Medium for normal work and Low for simple tasks. I have probably used High only 3-4 times in total. I always clear context or use /compact when old context is no longer relevant. I genuinely do not know what else I am supposed to do to preserve my limits.
I understand that usage is not measured by lines of code or wall-clock time alone. Context size, reasoning, tool calls, subagents, and output all matter. But that is also the problem: the practical cost of a task has become extremely difficult to predict.
I see people with $100 and $200 subscriptions saying they burn through their limits in 2-3 days. Tibo’s resets were the only thing saving them. Is that really considered normal now?
People are already making X skills to reduce usage. sol-advisor, for example, delegates work to Luna. Yes, that makes limits drain more slowly, roughly at the level they did before GPT-5.6. But what is the actual answer here? Are we supposed to use a smaller model just to get back to the old usage level?
For me, that feels either like a very transparent “pay more” message or like a seriously bad release. There were also plenty of API and Codex issues at launch.
Luna is not a bad model in general, but it does not work well for the coding tasks I use it for. It is only usable on Max or xHigh reasoning, which makes it incredibly slow. Its knowledge is not enough for me to rely on it the way I used to rely on GPT-5.5. It often writes overly complicated, poor code and agrees with almost any questionable approach.
In one case, Luna Max, launched through sol-advisor with Sol High, wrote around 300 lines of code across 12 files, plus 7 additional empty files. The entire stack still consumed about 9% of my weekly limit.
I also tried using Opus 5 High for an entire day. I did not hit the 5-hour limit even once. I know a group of five people who share a single $100 subscription to Claude. They have never hit the weekly limit and only hit the 5-hour limit a couple of times. At the end of each weekly reset, they still have more than 40% left.
Based on my own usage, I strongly suspect that Sol Low would use their weekly limit much faster.
All the posts saying, “I am a Claude user, convince me to switch to Codex,” followed by replies about large limits and Tibo resets feel almost funny now. The resets will not last forever, and the large limits seem to be gone already.
Please, if you have the same problem, like or reply to this post. And if you are not seeing it, please share your own observations.
I really need to know whether I am losing my mind here 
