OpenAI Fixes Codex After Developers Burn Through Weekly Limits In 2 Days

OpenAI said it fixed a Codex usage-limit problem after some users burned through coding credits faster than expected.

Key Points:

Codex Limits

Thibault Sottiaux, engineering lead for Codex, said late Monday on X that the coding agent had consumed more compute than intended because some background systems were doing extra work.

Users complained over the weekend that similar tasks were draining limits faster than they had one week earlier, including on higher-priced plans. Sottiaux said OpenAI opened a Sunday “warroom” to review the reports and reset usage caps across the service.

Codex usage limits show how much compute an AI coding task has used, with the dashboard displaying the figure as a percentage. More complex tasks consume credits faster, and available usage varies by subscription tier.

Sottiaux said auto-review, which checks code without direct human action, and helper “subagents” sometimes ran more often than intended. In some cases, those systems ran twice or retried too aggressively after errors.

The dashboard also showed activity that was not actually charged to users, he said. “All fixes are now deployed, and we've added more detailed monitoring so we can detect background-usage regressions sooner. We'll continue watching the results closely,” Sottiaux wrote.

Also Read: Crypto Bear Markets End When 5 Forces Align, Fidelity Says

AI Coding

The issue mattered because coding is one of the most compute-heavy AI tasks, and flawed accounting can make a paid plan feel unreliable.

A software engineer named Adam wrote Sunday on X that his $200 plan had previously taken a full week of heavy work to exhaust. He said the same plan burned through a week of usage in one day during each of the prior two days.

Srinivas Pendela, a developer, remained unconvinced Tuesday, writing that he had stopped using Codex for several days because the depletion was “brutal.”

The complaints landed in a market where OpenAI and Anthropic see coding agents as one of the clearest paid uses for AI.

Codex had already faced an outage earlier this month, while Anthropic has had similar pressure around Claude. In March, Anthropic adjusted Claude usage caps during peak hours as demand strained compute capacity, a reminder that AI coding tools have moved away from unlimited use as adoption has grown.

Read Next: Solana’s 4.51M Address Spike Raises One Question About Real Demand