Some OpenAI Codex users are raising concerns about how quickly their weekly usage allowances are being consumed, with developers calling for clearer information about exactly how individual coding tasks affect their limits.
The complaints do not prove that OpenAI has secretly reduced Codex allowances. However, reports posted on OpenAI’s Developer Community forum and the official OpenAI Codex GitHub repository show that some paying users are finding it difficult to predict how much work their subscriptions will cover.
One response to the issue is NerfTrack, an open-source desktop application created by developer Ayaan Lashari.
According to the project’s GitHub repository, NerfTrack is a local Tauri-based app that reads Codex usage records and estimates their API-equivalent weekly value.
The tool does not prove that OpenAI has reduced Codex quotas. Instead, it gives users another way to examine their own usage and estimate how much value their Codex activity represents.
OpenAI also publishes a token-based Codex rate card that explains how additional credits are consumed based on input, cached input, and output tokens.
OpenAI’s official Codex pricing documentation provides estimated usage ranges for five-hour windows and explains that more complicated tasks can consume limits faster.
The company also states that additional weekly limits may apply. Users can check their current allowance through the Codex Usage Dashboard or use the /status command during an active Codex CLI session.
However, OpenAI does not publish one fixed weekly allowance that applies to every user and workload.
That lack of a simple task-by-task breakdown has become one of the main complaints among developers.
One of the more detailed reports appeared in GitHub issue #36468 on the official openai/codex repository.
A ChatGPT Pro 20x user posting as FlowTether said their weekly allowance reset to 100% on August 1 before falling to 91% during a single GPT-5.6 Sol engineering session lasting roughly 4.5 hours.
The user also said the previous weekly allowance had been nearly exhausted within around one to one-and-a-half days and that they purchased another $40 in usage to continue working.
Another user, zkhrvv, reported a similar concern in GitHub issue #36488.
According to the report, 75.25 million tokens consumed around 26% of the user’s weekly allowance shortly after the August 1 reset. The user said about 96% of those tokens were cached reads and estimated that the same workload appeared to provide significantly less effective capacity than before.
A third report appeared in GitHub issue #36481, where Plus user yanse216 said their weekly usage indicator increased from 16% to 75% during five hours despite their local logs showing no model requests during that time.
These are individual user reports and do not establish that OpenAI deliberately reduced Codex limits.
Complaints have also appeared on OpenAI’s Developer Community forum.
In an April discussion titled “Understanding the New Codex Limit System After the April 9 Update,” forum user ZachA said asking Codex to summarize one method consumed 6% of their five-hour allowance. A follow-up request to modify the same method reportedly consumed another 5%.
The user said this occurred in a relatively small project and argued that the rate of consumption made sustained development difficult.
Another forum user, DerDerErIst, reported that a single large prompt consumed around 50% of their five-hour allowance and approximately 20% of their weekly allowance. The user said the five-hour limit was reached after three substantial prompts.
A separate user, shasixz, said one basic prompt consumed roughly 22 credits and argued that even relatively small development tasks were using paid credits faster than expected.
Another detailed complaint came from OpenAI Developer Community user D_Anil_Reddy in July.
The user said they had upgraded from ChatGPT Plus to Pro 5x and later Pro 20x while using Codex for professional software engineering.
Despite those upgrades, they said they exhausted both their weekly Codex allowance and GPT-5.3-Codex-Spark allowance sooner than expected.
The user also reported purchasing around Rs. 2,000 worth of additional Codex credits, which they said were consumed within only a few substantial engineering prompts.
OpenAI Support responded on July 23 that a complete per-task usage breakdown did not appear to be available. Support instead pointed the user toward the Codex Usage Dashboard and the /status command and said the request for more detailed task-level usage information would be passed along.
OpenAI’s own Codex documentation shows that usage can vary significantly depending on the model and workload.
More demanding models, larger contexts, longer outputs, tool calls, and additional agents can all increase resource consumption.
This means two prompts do not necessarily consume the same amount of a user’s allowance, even if they appear similar in length.
OpenAI now provides more information about Codex consumption through its pricing pages, rate cards, and usage dashboard.
Still, the reports posted on GitHub and OpenAI’s own developer forum show that some users want a much clearer breakdown showing exactly how much each Codex task consumes and why.
Get the latest news, tech updates, telecom insights, and business stories wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.
Technology and Automotive Specialist covering the latest cars, smartphones, AI breakthroughs, and...
Shares
