Codex CLI: Plus users hitting usage limits extremely quickly compared to competitors

Allow us to use gpt-5-mini in Codex CLI with ChatGPT Plus subscription, gpt-5 hit the usage limit really fast.

Edit: Why am I getting dislike and meh ? GPT-5-Mini only available for pay-as-you-go apiKey, not with ChatGPT Plus subcription. Trying to get more usages out of this.

This was new today me me - It has been 15 hours and my Plus usage hasn't reset, 1 or 2 prompt would be an improvement for me right now.

Plese make a transparent usage statistic and when to expect to reset them. We clearly need more time to use.

Same I've been blocked for more than 24h now :'(

For example, Gemini CLI gives you 1,000 queries per day — for free.

I also hit the weekly limit too quickly on Plus. I’ve only been using it for about 3 days, and I never even once hit the 5-hour session limit. But now I can’t use it anymore because the weekly limit was reached (since 5 hours total already passed). So I never actually used a full 5-hour slot, but it still counts as too much for the week. I think that’s too harsh.

We also need more statistics - like when exactly the limit will be lifted - so users can plan their usage better.

I just hit the limit according to CLI..but I can still use the web version. It seems to me that CLI and web actually have seperate limits. This needs to be clarified in the readme or better: CLI should share the limit of the web version.

okay...this is weird. now I can use the cli again...no "ran into limit" message anymore.

I've been using gpt-5-high with codex and the Plus plan, I hit the usage limits after about ~1.6M tokens. I assume gpt-5-high might contribute to hitting the limits faster, but I haven't tested it to confirm.

Yep after 4 - 5 hour limits this week. I have hit my weekly limit. Not that I used it for the full 5 hours of those sessions. Glad this was just a test to see if this would work and have other options.

Allow us to use gpt-5-mini in Codex CLI with ChatGPT Plus subscription, gpt-5 hit the usage limit really fast. Edit: Why am I getting dislike and meh ? GPT-5-Mini only available for pay-as-you-go apiKey, not with ChatGPT Plus subcription. Trying to get more usages out of this.

From what I saw in the /model selection you can choose gpt-5 minimal.

the thing about the modes below medium: they make codex WAY to dumb by comparison. It actually needs to work MORE to accomplish stuff that way, and needs more user guidance for it. this totally negates the "advantage" that the lower reasoning modes need less tokens.

The limits don't bother me, they're lower than Claude Pro plan but I'm still getting good value, but even more important than that a way that it can switch to my API key when limits are hit would be great, and documentation would also be great.

> Allow us to use gpt-5-mini in Codex CLI with ChatGPT Plus subscription, gpt-5 hit the usage limit really fast. > Edit: Why am I getting dislike and meh ? GPT-5-Mini only available for pay-as-you-go apiKey, not with ChatGPT Plus subcription. Trying to get more usages out of this. From what I saw in the /model selection you can choose gpt-5 minimal.

GPT-5 Minimal is just GPT-5 with very low reasoning effort (they changed “none” to “minimal” after I commented on this). It’s not the same as GPT-5-Mini. GPT-5-Mini-High is really good, and it’s also very cheap. I still don’t understand why I’m getting dislikes for asking for another solution to increase usage.

before latest update v0.21 it was really hard for me to hit the limits, after in 30 minutes!
I used it a week before updates hardly and it hit limits not so often as a claude code, now i am usinf claude code only, cause GPT on Plus is always in limits.

I can prompt all day in Cursor for $20 a month, why limit is so low in codex?

Hit a limit, don't even know how much I've used or how long it is going to take to reset the limit. /status-limits is game changer here!

I hit the limit on PRO and its not an 5hr block. It is a bug?

Most likely, because I can still use the chat interface

2025년 8월 22일 (금) 10:53, prime9dev @.***>님이 작성:

prime9dev left a comment (openai/codex#2448) <https://github.com/openai/codex/issues/2448#issuecomment-3215160176> I hit the limit on PRO and its not an 5hr block. It is a bug? — Reply to this email directly, view it on GitHub <https://github.com/openai/codex/issues/2448#issuecomment-3215160176>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAWKS6SM7ZCZ543NR2AO5QD3O5KKPAVCNFSM6AAAAACEHGNH6GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEMJVGE3DAMJXGY> . You are receiving this because you commented.Message ID: @.***>

Last week I hit the weekly limit unusually fast on Pro. After today’s reset, I ran a few checks.

In my case, this appears to have been caused by MCPs. With MCP tools connected, cached tokens were <1% of total. After removing all MCP connections, cached tokens increased and total token usage dropped by about 90%.

Usage with MCP:
📊 Token Usage
• Input: 816664 (+ 15104 cached)
• Output: 9029
• Total: 825693

Without MCP:
📊 Token Usage
• Input: 31834 (+ 432768 cached)
• Output: 7021
• Total: 38855

If your setup uses MCP, try removing MCP and compare token usage.
(More details: https://github.com/openai/codex/issues/2610)

here it is my only session Token Usage after which i got usage limit

📊 Token Usage
• Input: 844571 (+ 11164672 cached)
• Output: 79277
• Total: 923848
🖐 You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing), or wait for limits to reset (every 5h and every week.)

i didn't even met a context limit
⏎ send Ctrl+J newline Ctrl+C quit 923848 tokens used 18% context left

Last week I hit the weekly limit unusually fast on Pro. After today’s reset, I ran a few checks. In my case, this appears to have been caused by MCPs. With MCP tools connected, cached tokens were <1% of total. After removing all MCP connections, cached tokens increased and total token usage dropped by about 90%. Usage with MCP: 📊 Token Usage • Input: 816664 (+ 15104 cached) • Output: 9029 • Total: 825693 Without MCP: 📊 Token Usage • Input: 31834 (+ 432768 cached) • Output: 7021 • Total: 38855 If your setup uses MCP, try removing MCP and compare token usage. (More details: #2610)

That's right, I'm thinking about it too, it looks like the MCP is the reason. But this did not happen before, on the previous version I could work for a whole day, problems started in 0.23.0.

Thanks for all of the feedback everybody. We are actively dialing in the rate limits. We recently raised them multiple times (example) and plan to start providing a lot more clarity around them. We hear you. Stay tuned and make sure to keep the CLI up to date!

I use Codex CLI with Context7 MCP and teams subscription. Reached my limit after 30-60 mins of work.

Thats not fair, when OpenAI said 3.000 requests per week.

(ChatGPT Plus plan)

I reached the limit after relatively minimal usage:

• Input: 97391 (+ 1526400 cached)
• Output: 15973
• Total: 113364

This is on GPT-5 Medium

Same issue here on plus. last week I had a few couple of hours sessions with fairly large code base without any problems. was hitting the 5 hour limits a few times until weekly limit hit.

SInce yesterday I am unable to do much with the codex cli as it hits limit very quickly. two messages this morning with only a few thousand tokens used. I waited again for limit to reset after 5 hours and now I could only send one message, again codex mostly read the context I was providing and then hit limit again.

From my perspective the 50% increase for Plus seems to have had the opposite effect of limiting me completely.

I must note that I upgraded the codex cli from 0.21.0 to 0.23.0 as well yesterday, not sure if that is part of my severe limitation now.

Just for note, I did /logout and /login to ensure fresh auth for codex cli.

Product feels completely unusable now.

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 have also observed an interesting issue regarding limits:

I have signed up for 2 separate ChatGPT Plus account. After AccountA has reached its limits, I switched to using AccountB, and I'm immediately limited on AccountB. This is after having an ~10h gap from last use for both accounts.

Messaging regarding multiple accounts and how their limits might affect one another should be made clear.

I hit the usage limits really fast as a Plus user. It feels noticeably quicker compared to my previous runs.

🧠 Model
• Name: gpt-5
• Provider: OpenAI
• Reasoning Effort: Medium
• Reasoning Summaries: Auto

📊 Token Usage
• Input: 185,953 (+779,776 cached)
• Output: 9,287
• Total: 195,240

When can I use codex again? I hit the limit more than 5 hours ago and it's still staying "🖐 You've hit your usage limit.". Please give us some transparency!!

i also hit the limits today way faster than yesterday, around the same amount of usage as @wochagonnadu, on the plus plan. i was able to fill a context window to 70-85% before in the last few days before i got hit by the limit.

@wochagonnadu You can clearly see that the input is huge compared to the output. Moreover, the cache shows that context duplication occurs, which I described in another topic. This is largely due to the CLI itself, buffering, and similar factors. As a result, the limit gets exhausted extremely quickly. For example, even if you only want to make a small change, you end up reading 40k tokens unnecessarily and paying as if it were 100k.

I suspect that if you checked the API, the cost in your case would be for over 700k tokens, even though the input was only 180k. And the problem could have been solved with just 20k tokens.”

They would have to drastically raise the limit — the context management itself is weak.

I think they increased it but it's still wayyy too little

I've just hit the limit after 2 days of using Codex in VS Code, and it tells me to try again in FOUR DAYS. What the hell, it's so scammy, four days of subscription lost?

Edit: seems it's just Codex on VS, I still could use Codex from their website! I don't know why such a harsh limit there and not on the site though...

Not sure if this exactly belongs in this thread, but my experience is extremely frustrating:

I’m on Tier 1 API access with a GPT-5 model, which supposedly gives me a 30,000 token per minute limit. Yet, when I run Codex CLI in agent mode, I hit that limit in exactly 3 minutes on a simple task of refactoring a single file. This makes the tool practically unusable for any real development work.

Compared to alternatives like Cursor, which allow uninterrupted coding sessions for hours, Codex CLI feels severely throttled and broken.

Is this the same limit issue discussed here, or is this a completely different problem? Either way, the current experience is unacceptable for professional use.

I'm just chiming in to echo the other comments. On plus, I was able to use codex cli longer on average than claude code at the same price tier up until recently. Now codex hits some limit (how was the expression of the limit not part of the MVP for rate limiting?) in much shorter time than claude code. I do have serena connected, if MCP connections really are playing a part in this issue. i used to be able to swap tools and cover gaps, but now codex runs out so fast I end up with hours of downtime unless I want to add yet another paid subscription from another provider.

<img width="952" height="737" alt="Image" src="https://github.com/user-attachments/assets/9e06d386-94e8-40f1-b9bd-daf8ad15a3fe" />
Codex is Unusable.

It's made worse by being seemingly impossible to leave Tier 1. I'm at about 11 days since I put in $100 and still on Tier 1.

But the whole session did not do anything useful! I spent 20$ to test it, and it only complaints about limits?

________________________________
Od: Adsq @.*>
Odesláno: sobota 6. září 2025 21:12
Komu: openai/codex @.*>
Kopie: Tomas Rampas @.>; Mention @.>
Předmět: Re: [openai/codex] Codex CLI: Plus users hitting usage limits extremely quickly compared to competitors (Issue #2448)

[https://avatars.githubusercontent.com/u/213518341?s=20&v=4]AdsQnn left a comment (openai/codex#2448)<https://github.com/openai/codex/issues/2448#issuecomment-3263059272>

@tomas-rampas<https://github.com/tomas-rampas>
In the message, you can see that you have around 27k tokens in context and you’re trying to send a task of around 10k+.
This adds up to over 30k, so the TPM will never allow it to go through — resulting in what’s known as a deadlock.

Essentially, both issues are connected because they stem from a mismatch between the context size and what the API can actually handle.
The point is that a request with the maximum context cannot exceed either the model’s maximum context window or the maximum TPM.
So, the context must be adjusted in such a way that, in a single request, it never exceeds one or the other — otherwise, sooner or later, you’ll inevitably get stuck.

—
Reply to this email directly, view it on GitHub<https://github.com/openai/codex/issues/2448#issuecomment-3263059272>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAHDFWEOTZXAYUSGCBODA7T3RMWYLAVCNFSM6AAAAACEHGNH6GVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTENRTGA2TSMRXGI>.
You are receiving this because you were mentioned.Message ID: @.***>

<img width="2364" height="130" alt="Image" src="https://github.com/user-attachments/assets/469fcc4b-c3d6-4330-9486-84d0d5097fa2" />

Waiting ~4 days is insane, I thought it worked like Claude Code that has 5 hours reset.

Hey guys! I'm also on the Plus tier. If you want to stretch your usage a bit, here's what works for me: For 1. conceptual planning stuff – use high reasoning effort, 2. when you need to generate tasks with checkboxes – go with medium, 3. and for task-by-task coding or checkbox follow-ups – keep it on low. Been doing this for a week and I literally never hit any limits. Hope this helps someone!

Yeah I just hit the weekly Limit for Codex via Cursor within 18 hours apparently... The tool is amazing but the limits are awful. I've been alternating between the different Modes, mostly using Medium for tasks, Low&Minimal for low complexity changes even.

<img width="787" height="194" alt="Image" src="https://github.com/user-attachments/assets/61c55ab9-bda4-4583-af23-d71b276fde1f" />

<img width="538" height="81" alt="Image" src="https://github.com/user-attachments/assets/a0a921f6-7aad-4587-8386-4745074c71d3" />

I am using Claude Code with Max 5X subscription and wanted to try codex with Plus subscription. I reached weekly limit at second day, without getting any 5 hour limit, any warnings, nothing at all. Total usage in this period, about 4 hours. I can still use Web but not codex. I think this is very awful compared to Claude Code where Web limits and CC limits are shared and more clear. Also I can definitely say Claude Code with Pro subscription gives more usage compared to Codex with Plus IN 5 HOUR LIMIT

I am on teams plan, working on a small project, the 2 day cooldown is hit just after 2 prompts. And i dont even is that regularly.

I am using Codex with a Plus subscription in VS Code. I sent two requests before I got this message. This is very much not fixed, this issue should be reopened.

<img width="861" height="95" alt="Image" src="https://github.com/user-attachments/assets/39003161-219c-4c8b-bf65-42b53f452d0e" />

Up i just hit another limit after 7 prompts LMAOOOO

after upgrade to 0.36 just about 3 prompts it hit the limit and it asked me to try again in 3 days is this normal??

after upgrade to 0.36 just about 3 prompts it hit the limit and it asked me to try again in 3 days is this normal??

same issue for me, was really surprising
I did choose the new codex model, maybe it has much lower limits

Same here

Same here
■ You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again in 1 day 20 hours 19 minutes.

Same
■ You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again in 5 days 17 hours 52
minutes.

lol, same. 3 request untill rate limit on plus

<img width="605" height="44" alt="Image" src="https://github.com/user-attachments/assets/82418a9b-55d7-4225-94f3-37c11bd4e5d1" />

I’m currently on the Teams plan through my university, and I’ve just hit my weekly usage limit. While I’ve been supplementing with other tools like Claude (which resets every 5 hours) and Gemini (offering 1,000 free requests), and even GLM’s affordable options, it’s been challenging to understand why the usage limits for OpenAI’s offerings are so restrictive by comparison.

<img width="687" height="105" alt="Image" src="https://github.com/user-attachments/assets/9e87b86c-f3ee-4357-b163-b4d3611d1ad3" />
Moderated usage on Business account. Hit 5 hrs, limit only ones. Now can't work for 4 days.
Very nice -__-

very very generous limit XD

You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again in 2 days 22 hours 47 minutes.

wrote an article to try and get around this, first limits were a few hours, then a week then 4 hours into the next week another week. worse still the limits raise each time, have around 5 more days till it is usable locally again, so we came up with this as a partial solution https://articles.akadata.ltd/openai-quietly-lowers-codex-usage-limit-and-how-to-run-it-locally/ < not advertising here, it is my own article.

<img width="1260" height="376" alt="Image" src="https://github.com/user-attachments/assets/6e509635-2e47-49fc-b5cd-38c41a91c162" />
A workaround to monitor how close I am to the limits

https://github.com/rubens-amaral/codex_usage_report

2 prompts and it reaches 5 hr limit, too bad

https://articles.akadata.ltd/openai-quietly-lowers-codex-usage-limit-and-how-to-run-it-locally/

Try and do a video tutorial on this.

Thank you