You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Please let us know when the quota is expanded for pro users and I'll try again. It seems I already used 1 week worth of inputs in 2 days.
Do you have an estimate of how many tokens it took to hit the 1 week limit? Pretty scary that 2 days of use can be allowed by 5h limits and then hit the weekly limit.
Please let us know when the quota is expanded for pro users and I'll try again. It seems I already used 1 week worth of inputs in 2 days.
Do you have an estimate of how many tokens it took to hit the 1 week limit? Pretty scary that 2 days of use can be allowed by 5h limits and then hit the weekly limit.
I don't know. I just ran 1 terminal all day on sat and sun doing different tasks, running most of the day. I occasionally had one more terminal open doing some debugging on some other tasks, but didn't run it all the time. By the end of the day on Sunday it says limit reached and it is still not letting me use it after about 20 hours, so I guess I am done for the week.
Similar experience here, seem to have hit the weekly usage limit on Pro in 2-3 days of moderate use. Sessions of Codex working serially on assigned tasks with very occasional one-offs in a second window.
Claude Max 20x has handled much more without complaint. GPT-5 is the better model but the limits are crippling.
Similar experience here, seem to have hit the weekly usage limit on Pro in 2-3 days of moderate use. Sessions of Codex working serially on assigned tasks with very occasional one-offs in a second window.
Claude Max 20x has handled much more without complaint. GPT-5 is the better model but the limits are crippling.
Yeah, that is why I said it should be about 10x the current limit for pro users, so we won't see any limit hits. When the limit hits the code is often in an inconsistent state and every time it takes me 1-2 hours of work to fix things.
Please let us know when the quota is expanded for pro users and I'll try again. It seems I already used 1 week worth of inputs in 2 days.
Do you have an estimate of how many tokens it took to hit the 1 week limit? Pretty scary that 2 days of use can be allowed by 5h limits and then hit the weekly limit.
I don't know. I just ran 1 terminal all day on sat and sun doing different tasks, running most of the day. I occasionally had one more terminal open doing some debugging on some other tasks, but didn't run it all the time. By the end of the day on Sunday it says limit reached and it is still not letting me use it after about 20 hours, so I guess I am done for the week.
A single agent running on a Pro account, usually with breaks for a few minutes until I come back and start it on the next task, is reaching the 5 hour limits usually within a couple hours of what I'd call moderate use. Comparatively, Claude Code is capable of running a subagent team with Opus/Sonnet 4 on the MAX plan all day. I prefer GPT-5 quite a lot already, but as mentioned above, the quick and constant bumping into usage limits for even what feels like moderate use are just crippling. Worse is having to restart Codex after every reset rather than continuing the chat. Disappointing for an "unlimited" plan.
Reacted by smat-dev, Doug Haber, Mehdi Ataei, Mike and Egor
Just shipped 2 fixes for folks hitting Codex CLI rate limits:
Reduced token usage by improving the cache hit rate a ton
Fixed how the CLI counts usage (update to v0.21)
Thanks for the feedback as we iterate. Keep it coming, especially if you're on Pro hitting limits!
Please report it to Alexander Embiricos if you keep hitting the limits on Pro. They're evaluating how much this cache change improved that, and will follow up depending on feedback!
Reacted by smat-dev, enprop, Mehdi Ataei and ishanray
Just shipped 2 fixes for folks hitting Codex CLI rate limits:
Reduced token usage by improving the cache hit rate a ton
Fixed how the CLI counts usage (update to v0.21)
Thanks for the feedback as we iterate. Keep it coming, especially if you're on Pro hitting limits!
Please report it to Alexander Embiricos if you keep hitting the limits on Pro. They're evaluating how much this cache change improved that, and will follow up depending on feedback!
Is he on Github? I don't use Twitter. @seratch Would you be able to help? How do I know:
When I'm about to use up my 5 hour limit
What are those limits?
How do I paste a picture when using the CLI?
Are there plans for a planning mode?
These are the main reasons why I am again considering going back to Claude Code as Codex CLI for now even though I really like Gpt 5 as a model.
We've made some further adjustments to the rate limits and plan to add more transparency over time. Hang in there as we dial in the right numbers but we want you to feel like you're getting a generous amount with your ChatGPT plans.
Having the exact same issue, feels like mine was nerfed heavily after the "50% increase for Plus", and i tried to send my first message this morning and already hit the rate limit. Not sure if its the weekly or just a big? Messaging is very poor.
I just bought the business plan ($25/m/seat), invited 3 other team members, and started coding. Ironically, because we ran into the same problem with the $200/seat plan from Anthropic for claude code. Bought the OpenAI team plan at the $25 per seat rate maybe 2 hours ago. I hit my 5 hour rate limit in ONE HOUR of making one change to our app, between meetings, and 25% of the WEEKLY LIMIT. So on the business plan "for startups and teams" we get 4 hours a week of light use, per person, non-interchangeable? And my only option is to pay 10x FOR ALL SEATS to the $200/m/seat license? That is brutal. It seems both Anthropic and OpenAI are now playing the bait-and-switch game with ratelimits. We set up our company around these tools at certain rates, only to be ratelimited into the enterprise plans (no idea how much that costs). And if ONE user wants increased rate limits, ALL users need it, unless you have enterprise, which is god-only-knows how expensive.
Activity
aehlke commentedon Aug 8, 2025
I've asked one of the leads about it here https://x.com/manabiSRS/status/1953882144591184318
leoisadev1 commentedon Aug 8, 2025
I got in contact with Alexander Embiricos, he says that they know there is an issue and are working on fixing it
Aze1338 commentedon Aug 8, 2025
i was able to use it for 3 hours today on pro subscription. now im using it with api only because im tired of their bs
leoisadev1 commentedon Aug 8, 2025
aehlke commentedon Aug 9, 2025
He updated to say they made a big bump for pro. Anyone verify that's the case now?
takemiyamakoto commentedon Aug 9, 2025
It's about 2x what it was. But it still needs about 10x more to be useful. So please consider giving at least 10x more to pro.
takemiyamakoto commentedon Aug 11, 2025
Please let us know when the quota is expanded for pro users and I'll try again. It seems I already used 1 week worth of inputs in 2 days.
aehlke commentedon Aug 11, 2025
Do you have an estimate of how many tokens it took to hit the 1 week limit? Pretty scary that 2 days of use can be allowed by 5h limits and then hit the weekly limit.
takemiyamakoto commentedon Aug 11, 2025
I don't know. I just ran 1 terminal all day on sat and sun doing different tasks, running most of the day. I occasionally had one more terminal open doing some debugging on some other tasks, but didn't run it all the time. By the end of the day on Sunday it says limit reached and it is still not letting me use it after about 20 hours, so I guess I am done for the week.
artimath commentedon Aug 12, 2025
@takemiyamakoto i had about the same exact experience and currently am usage limited for the last 36hrs and counting.
smat-dev commentedon Aug 12, 2025
Similar experience here, seem to have hit the weekly usage limit on Pro in 2-3 days of moderate use. Sessions of Codex working serially on assigned tasks with very occasional one-offs in a second window.
Claude Max 20x has handled much more without complaint. GPT-5 is the better model but the limits are crippling.
takemiyamakoto commentedon Aug 12, 2025
Yeah, that is why I said it should be about 10x the current limit for pro users, so we won't see any limit hits. When the limit hits the code is often in an inconsistent state and every time it takes me 1-2 hours of work to fix things.
noname-oni commentedon Aug 12, 2025
Same on Plus.
rbergman commentedon Aug 12, 2025
A single agent running on a Pro account, usually with breaks for a few minutes until I come back and start it on the next task, is reaching the 5 hour limits usually within a couple hours of what I'd call moderate use. Comparatively, Claude Code is capable of running a subagent team with Opus/Sonnet 4 on the MAX plan all day. I prefer GPT-5 quite a lot already, but as mentioned above, the quick and constant bumping into usage limits for even what feels like moderate use are just crippling. Worse is having to restart Codex after every reset rather than continuing the chat. Disappointing for an "unlimited" plan.
aehlke commentedon Aug 12, 2025
New update: https://x.com/embirico/status/1955344043740893617
Please report it to Alexander Embiricos if you keep hitting the limits on Pro. They're evaluating how much this cache change improved that, and will follow up depending on feedback!
whoisyurii commentedon Aug 18, 2025
What about limits on web-version of codex? I've already used like 60-70 requests in 5 days.
ishanray commentedon Aug 18, 2025
Is he on Github? I don't use Twitter. @seratch Would you be able to help? How do I know:
These are the main reasons why I am again considering going back to Claude Code as Codex CLI for now even though I really like Gpt 5 as a model.
aehlke commentedon Aug 18, 2025
Yes as @embirico but he only solicited those reports on Twitter and I haven't seen him chat in here
Limit details in Codex are going to come after they settle rate limiting
Picture PR is WIP
No word on planning mode
gpeal commentedon Aug 23, 2025
We've made some further adjustments to the rate limits and plan to add more transparency over time. Hang in there as we dial in the right numbers but we want you to feel like you're getting a generous amount with your ChatGPT plans.
Here is an announcement about bumping it 50% for Plus a couple of days ago.
https://x.com/embirico/status/1959057942445269141
helpimfalling commentedon Aug 24, 2025
Having the exact same issue, feels like mine was nerfed heavily after the "50% increase for Plus", and i tried to send my first message this morning and already hit the rate limit. Not sure if its the weekly or just a big? Messaging is very poor.
noname-oni commentedon Aug 25, 2025
Absolutely the same for me. The previous week was okay, and this week, after "50% increase for Plus," I'm facing rate limiting in 3 messages.
utkusen commentedon Sep 3, 2025
I'm using API mode. Rate limited in 3 minutes I'm not joking. I thought I could drain my balance but that's not the case
Gian44 commentedon Sep 11, 2025
Hello. Where do I see the current rate I can use in Codex? I can't seem to find it anywhere
junaid1460 commentedon Sep 13, 2025
Literally got a message saying try after 5 days
ALeslie743 commentedon Oct 21, 2025
I just bought the business plan ($25/m/seat), invited 3 other team members, and started coding. Ironically, because we ran into the same problem with the $200/seat plan from Anthropic for claude code. Bought the OpenAI team plan at the $25 per seat rate maybe 2 hours ago. I hit my 5 hour rate limit in ONE HOUR of making one change to our app, between meetings, and 25% of the WEEKLY LIMIT. So on the business plan "for startups and teams" we get 4 hours a week of light use, per person, non-interchangeable? And my only option is to pay 10x FOR ALL SEATS to the $200/m/seat license? That is brutal. It seems both Anthropic and OpenAI are now playing the bait-and-switch game with ratelimits. We set up our company around these tools at certain rates, only to be ratelimited into the enterprise plans (no idea how much that costs). And if ONE user wants increased rate limits, ALL users need it, unless you have enterprise, which is god-only-knows how expensive.
aehlke commentedon Oct 21, 2025
@ALeslie743 this tool is really only usable beyond light experimentation with the $200/mo plan. I agree the business plan economics sound suspect.