5-hour limit on /fast mode

Open 💬 6 comments Opened Jul 1, 2026 by mobiletoly
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

While I understand that /fast mode burns tokens twice as fast from weekly quota, but what is the point to also burn 5-hr limit twice as fast? Because it is not very logical, I could do a work faster, but then have to sit and wait for 5-hr limit reset?

View original on GitHub ↗

6 Comments

github-actions[bot] contributor · 19 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #30212
  • #30785

Powered by Codex Action

Mahkhmood9 · 19 days ago

it takes 2x your usage,

Whats the point? it depend on your use case, if you are not a heavy user - you may want the 5-hours limit

1., heavy user
2., if you are using x-high or high , and you might calculate it for a small task if it worth it or not

but honestly , just use gpt 5.5 medium , anything above that makes it overkill imo

if you not , its by design for you to choosse not

Im on on the plus version , and its literally all you can eat baffe - honestly - i can use it for like 4 projects daily , I push about 40 prs a day in a very code bases

Mahkhmood9 · 19 days ago

But again as per , if you argue

im a codex user I use the codex for !8 hours a work day
this mean you want to make actions with it across your working day

also medium is very good -- for me its almost seems smarter than high or xhigh

IF you are having issues

the 5 hours limit sounds very reasonable, like 8 hours work day, 4 hours meeting ~ plus or minus; it vary
you should be calibaried with your "tokens bet"(new term coined here) , whats your end goal? what do you want to happen --- is what should always be in your mind before your prompt and ask it (its not an easy thing --- a lot of your prompts are a part of learning and expirement --- so you will change your prompts in the future (but the newer models ,,, are made for past you --- or the new users who try it just now))

but lets be fair openai , has a tiered system, do you need PLUS or PRO ?
---
I argue 2 things
either the current systme is okay

or 1. 20 or 50 minutes time out

  1. maybe reward the best users --- what can you do - if there is certain TOPICS or PROJECTs that people incinerate sacrificing a lot of compute and electron for nothing - then - you should use the telemtry you collected - and consider what to do --- do you force upon them a room clean up ? do you peacefully suggest them to make some changes (but lets be honest we shouldnt do it with people who maintain huge JAVA PROJECTS - with multi year long projects nogged and hatched -- a lot of times java project are poorly implemented -- while they could literally asked the computer to port it into something better [dont forget to say "prevent link rot"] )
  2. prompt based system (and forcing the developers to view this as the gold stands --> causing the model developers "eat it up" e g , repeatably maybe they want to simplify the project in the middle or whatever ---- not everything is eval - sometimes is the large bloated code bases --- and in learning - it also know what files not to read --- just the other day I had - (one of the biggest open source models out there , on opencode ) try to read and print .pyc files [lol] )

in the mission to cap users , which consume a lot - yet there is no moral right to over feed them despite their heavy use (e g they only make web - developemtn [maybe there is older sloppfied code and its afraid to change it --- LLM FEAR, LLM IGNORE , LLM SKIPS ---] OR they have large code bases [Answer to that] )

and also in terms of data , how you gonna train ai llm models -- if a lot of users - going to be in the belief they are correct about a past memory and argue (or it did not mention to the users some it believes --- or in grammer or slangs the users like --- sometimes people can real perfectally simplified american english --- and they very different ideas ---- people natrually like to disagree - but llm creators are tasked with bypass human errror and represnt the users with power and skill and speed an and knowledge -- that the users is not showen to BUT there is truth to it - because we all seen ai , llm make up stuff - so are just already used to fear and argue with it --- whe nits perfectlaly correct --- or we need to ask it the currect way)

what kind of developer are you?
the whom , see the file? and think of the functions ---
or you think for functions that can be in the file?

you understand where is it going? what i mean?

again

  1. x-high or high == can become faster (should i use it? ---> is it worth the bet?)
  2. you are not a heavy user -
  3. u only use for few task you are a heavy user -- but you rely on yourself----
  4. You are a heavy users ,, but you only use llm for code review ,,, or-but you use it find parts of the code - or explain things to you (and you want it instant)

and if its you are just fine with the "non-fast" model , thats good! I also think gpt5.5 is BLAZING FAST

mobiletoly · 19 days ago

but honestly , just use gpt 5.5 medium , anything above that makes it overkill imo - no, using high or xhigh is not overkill at all, it really depends on the project and complexity

amarbunty · 18 days ago

from codex cli update @0.142.4 / 0.142.5 tokens are depleting faster like Titanic ship is sinking in the sea due to bottom hole.

Mahkhmood9 · 18 days ago
but honestly , just use gpt 5.5 medium , anything above that makes it overkill imo - no, using high or xhigh is not overkill at all, it really depends on the project and complexity

I dont think you gave it a try - most of programmings is attaching datapoints -- not considering 90 logic gates-- in most cases gpt 5.5 medium does the job perfectally