OpenAI Fixes Codex Quota Overspend Bug, Resets All Account Limits Due to Abnormally High Cache Hit Rate

According to Ducklings Intelligence monitoring, Thibault Sottiaux, the Engineering Lead of the OpenAI Codex team, announced that the underlying issue causing rapid depletion of Codex's AI programming system usage quota has been identified and fixed. He also mentioned that all affected accounts' quota limits have been reset.

Sottiaux explained that the abnormal quota consumption was triggered by performance optimization measures deployed earlier. During the rollback of these optimizations, the team discovered a significant drop in cache hit rate during long-session context compression compaction. When the system attempted to compress the history of long sessions to free up space, the cache misses caused Codex to recalculate and process the entire historical context, leading to a sudden loss of Token quota.

Currently, OpenAI has removed the faulty optimization code, and the system is back to normal operation. As users' quotas were unexpectedly deducted due to the system flaw, the official team has reset the Codex usage quotas for all accounts as compensation for the affected users.