OpenAI announced a usage reset for all paid Codex and ChatGPT Work users, along with fixes for several issues that caused excessive quota consumption, such as context compaction, memory, and goal instructions. After optimization, users can expect their usage to last 10-50% longer, and an in-app usage breakdown feature is planned.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
OpenAI resets usage credits for all paid Codex and ChatGPT Work users and fixes multiple underlying issues causing excessive token consumption, with expected credit durability improvements of 10% to 50%.
OpenAI's product lead Tibo announced on X that usage credits are being reset for all paid Codex and ChatGPT Work users. The team reviewed thousands of user reports and released a series of fixes. Depending on usage habits, users will find credits more 'durable' than before, with usage duration increasing by approximately 10% to 50%.
Compaction issue: During compression, old images were incorrectly retained, causing the context to be too large and triggering compression again. After the fix, heavy image users see about a 10% reduction in credit consumption.
Memory issue: Background memory processes inherited a 'stop hook' causing idle spinning, affecting less than 1% of users, but in extreme cases a thread looped 15,000 times checking if it could stop.
Goals issue: The `/goal` command did not stop after completion, or the model stubbornly retried ineffective tools indefinitely, wasting 15% to 70% of users' weekly credits in extreme cases.
Automations issue: Some custom scheduled tasks ran more frequently than configured.
Subagents issue: Smaller models (like Luna) called more capable (and more expensive) assistant models without explicit instruction; when the orchestrating model did not enable `/fast` mode, it incorrectly required subagents to run in `/fast` mode.
Computer History issue: Legacy code caused the system to repeatedly summarize overlapping activity records, consuming up to one-fifth of weekly credits in some cases.
Rolling task summaries issue: Normal conversation turns triggered extra background requests, increasing token consumption by about 1%, now disabled.
MCP issue: Some tool results were double-encoded, and some tool instructions were accidentally truncated, leading to re-fetching.
To prevent recurrence, the team made underlying architectural adjustments and set up alert mechanisms. Additionally, a new feature is in development that will allow users to view credit consumption details directly in the app.
This report is based on a tweet by OpenAI's product lead Tibo, reposted by Baoyu on X. All fix details and figures come from Tibo's original statement, which is a source claim and has not been independently verified.
OpenAI has significantly improved Codex credit durability by resetting credits and fixing multiple hidden consumption issues, but users should monitor whether actual results match official claims.
Primary report
Primary source