Usage limit on ChatGPT pro-plan single-threaded usage

Additional information:
I had been using Codex CLI (v0.16) on a Plus plan for some time. At some point, I received the message:

You've hit your usage limit

I then upgraded my subscription to Pro and continued working - Codex started working again. However, after 1–2 hours it failed once more. At that time, /status still showed:

👤 Account
• Signed in with ChatGPT
• Login: ***
• Plan: Plus

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

📊 Token Usage
• Input: 1,489,763
• Output: 5,133
• Total: 1,494,896

After upgrading Codex CLI, logging out, and logging back in, /status correctly shows Plan: Pro and zeroed usage:

👤 Account
• Signed in with ChatGPT
• Login: ***
• Plan: Pro

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

📊 Token Usage
• Input: 0
• Output: 0
• Total: 0

However, the usage limit is still being enforced.

I suspect this might be due to a limiter window that persists when upgrading the plan mid-window (e.g., during a 5-hour guardrail window), or an issue with plan caching affecting how limits are calculated after an upgrade.

I am also getting this same behavior and was surprised that just having it write some test cases and updating some code I hit usage limits with a Pro plan. I just upgraded to 0.19 so I will see if it's still happening.

Additional context on the issue:

This case is not about a marathon workload - it was a single small bug fix in a small repo, with total wall-clock time across both sessions under 1 hour. Still, the 5h/weekly usage guardrail triggered.

Timeline:

1. Session 1 - Pro plan (ChatGPT login), single-threaded, Reasoning Effort: Medium, model gpt-5.

• Debugging a frontend via Playwright, with the same error message continuously spamming the console logs.
• Ran for ~4,078,566 total tokens (input 4,066,978 [+801 cached], output 11,588; reasoning 8,384).
• Failed with:

🖐 stream disconnected before completion: stream closed before response.completed

1. Session 2 - Immediately restarted the same small fix task.

• Also debugging the frontend via Playwright, same repetitive console logs.
• Ran for a while, then failed with:

🖐 You've hit your usage limit. Limits reset every 5h and every week.

• /status at this moment:

Plan: Pro
Model: gpt-5
Reasoning Effort: Medium
Reasoning Summaries: Auto
Token Usage: total=3,382,096 (input=3,372,183, output=9,913)

Observations / concerns:

• The total real work time for the system was <1h, single user, one process.
• Both sessions were focused on debugging the same frontend issue via Playwright, with large volumes of repeated console log output from the same error.
• The combined ~7M token context may have been caused by the model attempting to fully read and process these repetitive logs.
• Hitting a 5h/weekly guardrail under these conditions doesn’t feel like “unlimited (subject to abuse guardrails)” for a Pro plan.
• It appears that an incomplete/disconnected stream from Session 1 still fully counted toward the guardrail window, so Session 2 quickly exhausted the remaining allowance in the same window.

Hypothesis:
This may be caused by:

• Guardrail accounting that fully charges token usage even if the stream ends before response.completed.
• No grace/retry credit when the failure cause is a transport disconnect or infra timeout.
• Potential runaway token consumption from unbounded tool output (e.g., reading entire repetitive Playwright logs into context).

Suggestions / requests:

1. Do not fully charge the guardrail window for incomplete/disconnected runs, or allow a single safe retry in that window without double-charging.
2. Provide clear reason + ETA to reset when a guardrail triggers (see my separate FR for global usage/reset info).
3. Revisit guardrail thresholds for ChatGPT-login on Pro, especially for single-threaded use.
4. Surface specific error type (infra timeout vs. limit) so users know how to mitigate (e.g., split into smaller steps, lower Reasoning Effort).
5. Consider adding configurable limits or filtering for tool output (e.g., max lines/tokens from Playwright logs) to prevent runaway context growth from repetitive logs.

The limits on the Pro plan are supposedly generous: https://x.com/embirico/status/1953590991870697896

But I've hit the 🖐 You've hit usage your usage limit. Limits reset every 5h and every week. wall after doing the same velocity of work as I do on Claude Code with 20x max plan.

Same here. I've decided to try GPT-5, bought a Pro plan for trying, and while in the Claude Code 20x Max plan, I rarely hit the limits (even after they've made them stricter), here I've hit those after a few hours of similar work.

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

📊 Token Usage
  • Input: 2515609 (+ 6306528 cached)
  • Output: 28287
  • Total: 2543896

That is single agent mode, I haven't tried multiple agents in parallel, but the project I've tried to work on is investigation heavy (requires to document a lot, create plans, then test a lot).

I'm having the same issue. I had a pro plan and coded for maybe 1-2 hours, not too many prompts, towards the end I had 2 threads going, but total token usage was probably under 4m, and I have a usage limit!?

anyone have a fix? I updated the package and have the same issue.

Same thing happening over here. Supposed to be unlimited, but hit the limit within an hour or so.

Also just ran into this after only a small amount of work in a single session. Also have been someone working with Claude Code on the 100/mo plan and never ran into limits here.

It's hard to believe I was at a real limit, especially given that my context wasn't even full:

<img width="736" height="162" alt="Image" src="https://github.com/user-attachments/assets/a72952e0-5dd1-418b-9bc4-f9b47d2c5c9f" />

I really want this to be a bug and not the limit of the normal Pro plan...

I really want this to be a bug and not the limit of the normal Pro plan...

That limit I saw when I've accidentally used free tier instead of paid - I've fixed that by logging out and logging in back again.

But the other limit that resets every 5h is also rather bad :(

Same issue here, made minor changes to two websites in two Codex CLI terminal windows at the same time. Used it for about 1 hour - hit usage limits.

Even doing moderate updates, I am reaching the maximum tokens per hour limit, making it unusable. I have to exit the cli and restart Codex. As it doesn't have the -c option like claude code, I'm constantly having to go back and remind it where we were at when it failed. I tried askihng it to put guardrails and delays between updates, but to no avail. Maybe this is a result of 5 being faster? Whatever the reason, the hard fail when it happens, rather than an elegant pause is the big issue here.

@Civil the codex lead tweeted that they raised the limits (looks about double) for pro tier after you posted that. curious if it's working well now. I'm hesitant to upgrade

@aehlke yeah, today I haven't hit those running window limits like before, so maybe they've fixed those. So far I'm playing around on a task that Claude Opus failed implement no matter how hard it tried - and so far GPT-5 do better overall, but it pauses a lot asking for directions and guidance, too much to my taste, and it also for some reason hesitate to run some of the commands... But I guess that is a story for another bug

@Civil are you trying thinking=high?

I'm so far on default medium thinking.

Here's a new data point:

Codex CLI (Pro)
2h 15min until 5h Rate Limit

2 sessions running concurrently
Near non-stop with immediate task addition

Total Token = 3,633,057 + 8,751,193 + 1,435,625 + 4,043,907 = 17,863,782

Source: https://x.com/y_esnya/status/1954445668283560154

This is a discount on API rates, but is not competitive for CLI coding agent subscriptions

<img width="952" height="545" alt="Image" src="https://github.com/user-attachments/assets/4911e346-6a75-47b6-b214-04e02fce25ef" /> I worked on the project for about three hours now, but the project wasn't finished, but despite it being a unique and complex project, GPT 5 worked very well and quickly. MVP is almost finished, but the real problem is that I didn't bother at all. I set up the template at the beginning, gave it a PRD, and just told it how to work with a good Prompt, and the project is almost finished.

🖐 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.).

Its more than 5 hours so how long do i have to wait more ?

@VimsRocz we don't have reporting yet to tell. but either there's a bug with how it counts 5 hours, or you've miscalculated how it works, or you've hit the weekly limit and need to wait another ~7 days.

@VimsRocz we don't have reporting yet to tell. but either there's a bug with how it counts 5 hours, or you've miscalculated how it works, or you've hit the weekly limit and need to wait another ~7 days.

7 days more from now ?

@VimsRocz either from now or from the first use in a 7 day window, likely the latter actually. But we don't know yet

I have pro and just asked it to check the history files in the project to catch up, 🖐 stream disconnected before completion: Rate limit reached for gpt-5 in organization org-xxx on tokens per min (TPM): Limit 30000, Used 17417, Requested 24901. Please try again in 24.636s. Visit https://platform.openai.com/account/rate-limits to learn more.
Set to GPT 5 and medium.

🖐 You've hit your usage limit.

With the Plus plan, the first two sessions were kind of okay, but then I started facing limits after several requests every 5 hours when the limits are refreshed. It seems like there is a weekly cap and a cap that refreshes every 5 hours. Hard to use the tool with such limits.

---

upd: after passing 5h, limits are not being refreshed at all

I've been using Codex on the Plus plan for 3 days (2-3 5 hours sessions per day) and now 🖐 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.). Is this a weekly limit hit? No info when it will reset. If this is a weekly limit already after 3 days then it's not a generous level to work with.

I'm on Plus plan, but I cannot even complete /init command with new projects. Also there is no way to resume after 5 hours, it's completely unusable for me. Sometimes it stops while writing to a file. In Claude Code at least i can just type continue and it uses the same session to continue where it cut off.

🖐 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.). Its more than 5 hours so how long do i have to wait more ?

me to i am team user and for several days i have been working, hit the rate limit -> waited 5 hrs -> work again
but today after 5 hrs its still saying
🧠 Model
• Output: 0
• Total: 0

user

🖐 You've hit your usage limit. Limits reset every 5h and every week.

I'm on Plus plan, but I cannot even complete /init command with new projects. Also there is no way to resume after 5 hours, it's completely unusable for me. Sometimes it stops while writing to a file. In Claude Code at least i can just type continue and it uses the same session to continue where it cut off.

Yeah I am having similar issues. Can't even get through a single prompt without it hitting limits after "thinking" for about 3-4 mins. I have both Teams and Plus, and same problem. Its really unusable and frustrating.

Pro plan, looks like got weekly ban. Total working time is about 15hr during the week, with regular rate limiting on a mid-size project.
With the current Pro advertising on a competitive market I would expect working 24/7 in a single thread.

Pro plan, looks like got weekly ban. Total working time is about 15hr during the week, with regular rate limiting on a mid-size project.

Please report this to https://x.com/embirico (when Twitter is back online) as he's been soliciting this feedback and guaranteeing rate limits won't be hit (under non-scripted single-agent use)

Context from a real feature build (session restore) showing Pro limits block single-thread work
Workload: single agent, single thread, medium repo.
Feature: session restore across TS CLI + Rust TUI; small diff but many touchpoints.
Size: ~70 KB total (incl. ~20 KB sample project).
Changes (external): unified action selector; server restore with fallback; viewer focus + full-history toggle; persisted meta; synthetic aborts for unfinished tool-calls.
Changes (internal): resume reconciliation in Rust core; additive JSON/JSONL formats; shared transcript renderers; optional segmented restore with token estimate.
Runtime: ~5 hours during a day (~12hrs real time, waiting for rate-limits ban); earlier in the week another ~7 hours on other tasks.
Limits: several 5-hour locks mid-task, and finally a weekly lock.
Counters at lock varied: at times /status ~264k total; other times planner input built up, e.g.
Input 17,022,022 (+216,064 cached) • Output 73,532 • Total 17,095,554 (one run near 30M input).

Result: could not complete even this mid-size feature.
Request: expose 5h/weekly counters + next reset, make gates backoff+resume instead of terminal, and document a path/tier for 24/7 single-thread on a medium repo.

@embirico tagging you for pro-tier rate limit reporting above (since X is down). thanks so much

@embirico tagging you for pro-tier rate limit reporting above (since X is down). thanks so much

please increase teams rate limits as well for single thread/non scripting chats

codex-cli 0.21.0

/status
📂 Workspace
  • Path: ~/XXXX
  • Approval Mode: on-request
  • Sandbox: workspace-write

👤 Account
  • Signed in with ChatGPT
  • Login: XXXX
  • Plan: Plus

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

📊 Token Usage
  • Input: 0
  • Output: 0
  • Total: 0

user
hi

🖐 To use Codex with your ChatGPT plan, upgrade to Plus: https://openai.com/chatgpt/pricing.

Am seeing this after sometime, am not sure what the limits are and why does it even show this, am confused.

> 🖐 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.). > It's been more than 5 hours — how much longer do I have to wait? Me too. I'm a team user and for several days I've been working, hit the rate limit -> waited 5 hrs -> worked again, but today after 5 hrs it's still saying 🧠 Model • Output: 0 • Total: 0 user 🖐 You've hit your usage limit. Limits reset every 5h and every week.

Hello, I'm also a team user. May I ask how much Codex usage you typically have when you encounter the 5-hour usage limit?

> > 🖐 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.). > > It's been more than 5 hours — how much longer do I have to wait? > > > Me too. I'm a team user and for several days I've been working, hit the rate limit -> waited 5 hrs -> worked again, but today after 5 hrs it's still saying 🧠 Model • Output: 0 • Total: 0 > user > 🖐 You've hit your usage limit. Limits reset every 5h and every week. Hello, I'm also a team user. May I ask how much Codex usage you typically have when you encounter the 5-hour usage limit?

yes right now i have had like 10 requests, model_reasoning_effort high and got hit the usage limit
• Plan: Team

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

📊 Token Usage
• Input: 1208148 (+ 7684864 cached)
• Output: 79820
• Total: 1287968

I was working on two different parts of my codebases with different GPT-5 instances on high reasoning. I'm on the GPT-5 Pro Plan, I wonder if concurrency triggers the guardrail limit. I come from being a Claude Code Max user (same $200/month) For me, working (concurrently) on different parts of a codebase is essential.
That said, maybe I simply used too many token within a 5 hour window. Would love the ability to prompt for a model change from high to minimal reasoning..which would probablt save quite a few tokens

Looking at my OAI $200/month Pro Plan deal says unlimited...maybe its a bug?

<img width="2978" height="1954" alt="Image" src="https://github.com/user-attachments/assets/4f51c91d-8fad-49e0-ab5c-7d742efa78c7" />

We have made (and continue to make) changes to rate limits. At the moment, they are much higher than they were when this was first created so I'm going to close this out for now.

We're on a business team plan and after 3 days we're hitting a usage limit. Using Codex. Is this expected and what are our options to raise the limit?

<img width="890" height="107" alt="Image" src="https://github.com/user-attachments/assets/25d777a3-17f0-4381-8ce7-fdcffd41e6e4" />

We're on a business team plan and after 3 days we're hitting a usage limit. Using Codex. Is this expected and what are our options to raise the limit? <img width="890" height="107" alt="Image" src="https://github.com/user-attachments/assets/25d777a3-17f0-4381-8ce7-fdcffd41e6e4" />

I don't know about the business tiers except that the regular team one is equivalent to Plus. Plus is only appropriate for light use and testing it out occasionally - like a couple good short sessions per week. If you are doing any level of serious work you must upgrade to Pro which gives around 10x more usage.