OpenAI restores Codex, ChatGPT Work limits after traffic spike

OpenAI has once again reset the usage limits for its Codex coding agent and the newly launched ChatGPT Work, which reflects the growing strain that AI agents have been creating on the computing infrastructure.

On July 11, Codex engineering lead Thibault Sottiaux took to X and announced that usage limits for both products would be restored completely to all users within about 30 minutes. He thanked the community for “pushing our systems to the absolute limit,” adding that OpenAI had “never seen traffic increase so quickly.”

Repetitive resets indicate a larger challenge that goes beyond momentary service issues. While OpenAI, Anthropic, and Microsoft deliver advanced workplace agents, they are also facing the high processing demand posed by such systems. These agents are much costlier to operate because they also conduct lengthier and more complex tasks compared to conventional chatbots.

On Friday, another reset was recorded just a day prior. On July 10, Sottiaux stated that OpenAI had already increased the limits of usage of Codex and ChatGPT Work while assuring of another reset related to the company’s rollout updates.

Hello beautiful people! We have reset usage limits across Codex and ChatGPT Work. And another one will come later in the day. Rejoice.

Now that I have your attention, a quick update on ChatGPT Work, Codex and all the updates we shared yesterday.

We’ve spent the last 24 hours…

— Tibo (@thsottiaux) July 10, 2026

The timing is worth mentioning since ChatGPT Work has only been recently launched. According to Fox Business, OpenAI launched ChatGPT Work on July 9, a GPT-5.6-based enterprise-focused agent that works in conjunction with workplace apps to generate reports, spreadsheets, presentations, and other business materials. The release of such a compute-intensive product at a time when the existing infrastructure is already under stress can help explain why users are facing repeated quota resets.

Inside the Codex bug

The new series of resets comes in the wake of troubles that came to light in the last week of June, when many paying users of Codex noticed that their credits were disappearing much faster than anticipated.

As Cryptopolitan reported, OpenAI attributed the problem to a malfunction in its fraud-prevention system that erroneously applied rate limits to certain accounts while using up developer credits. Some users claimed that the rate of credit consumption climbed by a factor of 10 to 20 times, while those on the $200-per-month Pro plan claimed that they watched about $40 in credits vanish in a few hours.

Sottiaux went on to explain that his team worked through the weekend in what he called a “war room” where they combed through the logs and found the problem. OpenAI made three resets of quotas from June 28-29 before making one more reset to clients who were affected.

As it turned out, there was no single reason for the incident. Business Insider noted that Codex was doing a lot more than anticipated. Automated code reviews, helper subagents, and retry mechanisms may have run several times after an error, consuming excess resources each time. At the same time, the usage dashboard showed an activity that was never even charged, thus adding to the mess.

“All fixes are now deployed,” Sottiaux said after the occurrence, adding that OpenAI has established monitoring systems in order to inform the company if there are any issues in the future. However, July’s latest resets suggest that while the bug has been fixed, capacity issues are persisting.

The whole sector is metering harder

OpenAI is not the only AI firm that has restricted access in recent times due to increased demand.

Earlier in the year, Anthropic lowered Claude usage caps while service was in high demand, and one issue with Claude back in March interrupted developers who had begun to rely on the assistant for programming tasks. The trend now across the industry is that companies are avoiding unlimited access to their services as demand for such has outgrown available computing power.

The study sheds light on the matter. In an April study authored by the Stanford Digital Economy Lab, researchers found that agentic coding jobs require approximately 1,000 times more tokens than conventional coding conversations, with input tokens being largely responsible for that cost. They also found that performing the same task several times can result in significant variation in token usage of up to 30 times despite the fact that the increased amount of tokens did not necessarily improve the result.

Those findings highlight why it remains problematic to charge for AI agents. Since the cost of computing different tasks can be so variable, providers have very few alternative ways other than usage caps, quotas, or periodic resets to control usage.

For developers, however, those measures bring a different set of problems. Many of them are now organizing their work based on quotas instead of deadlines. OpenAI says it will continue monitoring usage and provide further updates if needed, but the repeated resets underscore a challenge facing the entire industry: today’s AI agents are becoming more capable faster than companies can build the infrastructure required to support them.