Codex fixes eight quota anomalies, freeing up to fifty percent more paid usage

Codex, the application programming interface unit of OpenAI, has resolved eight separate quota anomalies that were causing unexpected usage spikes for paying customers. The fixes reset usage limits so that the same paid quota now supports between ten and fifty percent more work, depending on how the service is used. OpenAI did not change the price of its paid tiers and said the changes apply automatically to existing subscriptions.

The anomalies affected a range of Codex behaviors, including background memory processes that could not be stopped, tasks that continued running past their intended end, and automatic overclocking that consumed extra resources. In one documented case, an anomalous goal task ran for a week and consumed fifteen to seventy percent of a single week’s quota. Another issue repeatedly summarized conversation history, adding up to one fifth of a quota each week in some scenarios. OpenAI said it will add new monitoring to alert engineers if similar regressions reappear and will show users a clearer breakdown of quota consumption inside the application.

The fixes come as OpenAI Codex expands beyond code generation into broader software workflows. The company has also reset quotas for ChatGPT Work users, who combine Codex with the ChatGPT interface for multi-step tasks. OpenAI did not disclose how many customers were affected by the anomalies or whether it will issue refunds or credits for past overages.

Codex is a paid API service that lets developers integrate OpenAI models into their own applications. It is used by startups and large companies to automate coding, testing, and documentation workflows. OpenAI did not announce any new pricing tiers or feature updates alongside the quota fixes.

Reported by ifanr.