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
I'm surprised, my usage style hasnt changed, but it seems that i have used 22% of weekly allotment in one day???? never has happened before, even on my biggest multi-threads + subagents days.
I had exactly the same feeling today. Yesterday evening I had more than 90%, today I had little use (I was expecting to be not lower than 70%) and find out to be at ~30%
We haven't made any changes to usage tracking or rate limits, so you must have used more tokens than you usually do.
I certainly believe this, but is there anyway to know if something has changed (no codex app changelogs is nerve-racking) and now my config, or tool calls or something is 'leaking' or maybe consuming duplicate, or unnecessary tokens?? how can i even investigate myself to see what the burn is coming from, so i can decide how to manage myself. that observability would be helpful, especially considering no insight into subagents, and where they at and what theyre doing in the background.
019c9791-29fd-74f2-b4d7-e004cc5636e5 (just ate another 7% of weekly in 30 minutes, 4 active sessions, no subagents that im aware of)
and overnight (in Europe) without any use at all, other 5% disappeared. I do not use any subagents, only one single session of codex cli. And I find rather strange that there are three independent issues opened here in the same time span of people complaining of the same unexpectedly fast token consumption. If you @etraut-openai say that you have not made any change, I totally trust you (1), but maybe there is a real issue affecting the measuring system.
(1) to be clear, I am very satisfied by open-ai service and very thankful for the great service. I am just reporting what I am convinced is a problem requiring attention from your side.
My guess is that it could be due to a recently introduced background terminal feature that the agent can monitor in real time. The agent wakes up about every 30 seconds to track the progress of running commands. Each time it wakes up, it probably sends the full session context, which could be draining the limits. However, I’ve been using this feature since it was experimental and haven’t seen any increase in recent usage.
I have updated to version v0.104.0 yesterday before I noticed this effect and I do not use anything experimental. In general I am never closing the session and do not have background running commands started by the agent, at least not that I know. And just now updated to v0.105.0, let's hope that it won't use all my remaining tokens :)
I've also noticed that the limits are being consumed faster than they were a few days ago. For example, a five-hour limit was almost depleted after five requests. Previously, it would have taken about 15% of the five-hour limit.
I've also encountered the same issue in recent days. Using codex cli & codex vscode extension, around 10k tokens can take up to 10% 5h usage limit. There might be a bug in the backend counting system somewhere.
Feedback has been sent, 019ca53d-ac3d-77a1-bcfc-6ff070a1b5c3
Also noticing my usage appears to be dropping faster than usual with similar ot lighter workload this weekend. CLI version 0.106.0. Pro plan. Weekly usage interestingly is also bouncing around, e.g. initially it's 62%, then a moment later it's 59%, then it's 62% again.
We haven't changed anything in terms of usage accounting or rate limits. We have rolled out various performance improvements that may result in faster usage of tokens, but the most likely explanation here is that your usage patterns have changed.
Reacted by Raxxoor, Almaz, David Dolph 🐺, Atis, Dmitry Murzin, cjstrads, eduardoadipe and mefree2098
where can i go to see and verify my increased token usage? -BenOn Mar 1, 2026, at 9:57 AM, Eric Traut ***@***.***> wrote:etraut-openai left a comment (openai/codex#12728)
We haven't changed anything in terms of usage accounting or rate limits. We have rolled out various performance improvements that may result in faster usage of tokens, but the most likely explanation here is that your usage patterns have changed.
—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>
Same issue here and am just using the codex app, not terminal. Using it for fairly basic work and at medium or low reasoning and it is chewing through the usage. I am using it less than a week ago but consuming much faster. Either the models are setup to consume more (more sub agents) or there is a bug, nothing has changed from my side. Concerning if OAI does not know what is going on.
Activity
human-bee commentedon Feb 25, 2026
019c922e-b020-7a60-8e12-28a2e7b004ba
github-actions commentedon Feb 25, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
etraut-openai commentedon Feb 25, 2026
We haven't made any changes to usage tracking or rate limits, so you must have used more tokens than you usually do.
abulgher commentedon Feb 25, 2026
I had exactly the same feeling today. Yesterday evening I had more than 90%, today I had little use (I was expecting to be not lower than 70%) and find out to be at ~30%
human-bee commentedon Feb 26, 2026
I certainly believe this, but is there anyway to know if something has changed (no codex app changelogs is nerve-racking) and now my config, or tool calls or something is 'leaking' or maybe consuming duplicate, or unnecessary tokens?? how can i even investigate myself to see what the burn is coming from, so i can decide how to manage myself. that observability would be helpful, especially considering no insight into subagents, and where they at and what theyre doing in the background.
019c9791-29fd-74f2-b4d7-e004cc5636e5 (just ate another 7% of weekly in 30 minutes, 4 active sessions, no subagents that im aware of)
abulgher commentedon Feb 26, 2026
and overnight (in Europe) without any use at all, other 5% disappeared. I do not use any subagents, only one single session of codex cli. And I find rather strange that there are three independent issues opened here in the same time span of people complaining of the same unexpectedly fast token consumption. If you @etraut-openai say that you have not made any change, I totally trust you (1), but maybe there is a real issue affecting the measuring system.
(1) to be clear, I am very satisfied by open-ai service and very thankful for the great service. I am just reporting what I am convinced is a problem requiring attention from your side.
DoubleBoba commentedon Feb 26, 2026
My guess is that it could be due to a recently introduced background terminal feature that the agent can monitor in real time. The agent wakes up about every 30 seconds to track the progress of running commands. Each time it wakes up, it probably sends the full session context, which could be draining the limits. However, I’ve been using this feature since it was experimental and haven’t seen any increase in recent usage.
abulgher commentedon Feb 26, 2026
I have updated to version v0.104.0 yesterday before I noticed this effect and I do not use anything experimental. In general I am never closing the session and do not have background running commands started by the agent, at least not that I know. And just now updated to v0.105.0, let's hope that it won't use all my remaining tokens :)
almaz1011996 commentedon Feb 28, 2026
I've also noticed that the limits are being consumed faster than they were a few days ago. For example, a five-hour limit was almost depleted after five requests. Previously, it would have taken about 15% of the five-hour limit.
bentz commentedon Feb 28, 2026
I've also encountered the same issue in recent days. Using codex cli & codex vscode extension, around 10k tokens can take up to 10% 5h usage limit. There might be a bug in the backend counting system somewhere.
Feedback has been sent, 019ca53d-ac3d-77a1-bcfc-6ff070a1b5c3
Ixe1 commentedon Mar 1, 2026
Also noticing my usage appears to be dropping faster than usual with similar ot lighter workload this weekend. CLI version 0.106.0. Pro plan. Weekly usage interestingly is also bouncing around, e.g. initially it's 62%, then a moment later it's 59%, then it's 62% again.
etraut-openai commentedon Mar 1, 2026
We haven't changed anything in terms of usage accounting or rate limits. We have rolled out various performance improvements that may result in faster usage of tokens, but the most likely explanation here is that your usage patterns have changed.
human-bee commentedon Mar 1, 2026
cjstrads commentedon Mar 4, 2026
Same issue here and am just using the codex app, not terminal. Using it for fairly basic work and at medium or low reasoning and it is chewing through the usage. I am using it less than a week ago but consuming much faster. Either the models are setup to consume more (more sub agents) or there is a bug, nothing has changed from my side. Concerning if OAI does not know what is going on.