OpenAI has identified three Codex-related sources of unexpected quota consumption after users reported that their allowances were depleting faster: inefficient repeated context compaction in long, image-heavy sessions, excessive use of Computer History in some high-consumption scenarios, and automatic session-title generation using more quota than expected. The company plans to push a fix on Sunday in the United States and fully reset Codex usage for all paid subscribers at about 2 p.m. Pacific time, or around 5 a.m. Beijing time on Aug. 24. The development revises an earlier explanation focused on sub2api activity and subscription sharing. It follows OpenAI's distribution of one-time reset vouchers to paid Codex and ChatGPT Work users after Codex surpassed 20 million active users, as well as earlier quota-related incidents involving background work, redundant sub-agents and increased tool calls.